With all the new AI models in Microsoft Copilot along with the first party agents that serve spefici functions, it can be confusing to know which tool to use for the right task. Microsoft have two specific agents within the Microsoft 365 Copilot domain Researcher and Agent designed to carry outr specific functions. We also now have GPT-5 (now also baked into Copilot) which brings the new “smart” mode, allowing it to switvch between languiage models based on the task at hand.
In this blog, I aim to break down the three tiers of Copilot capability— GPT-5 Chat, GPT-5 Reasoning (also known as Smart mode), and Researcher and to look at the differences, similarities and what is best for what!
GPT-5 Chat: Fast, Friendly, and to the Point
Best for:
General Chat and Quick Q&A
Content Summarisation
Creative brainstorming
How it thinks: Chat mode is is your rapid-fire assistant. It delivers supoer quick responses to youur questions without deep analysis. This is ideal for when you need an answer or exmplanation, clarity on something and anything not requiring deep thinking, reasoning or complexity.
Example prompt:
“Write a short social post abou the latest updates to [product].”
When to use it: When you want brevity, speed, and a touch of flair. Think elevator pitches, tweet-length summaries, or light creative riffs.
GPT-5 Reasoning: Structured Thinking for Strategic Tasks
Best for:
Logical analysis
Problem-solving
Multi-step planning such as building a travel itinery, project plan or talk track
How it thinks: This mode (which GPT-5 will switch too automatically if needed), rolls up its sleeves and gets analytical. It’s designed for tasks that require synthesis, deep exploratory analysis, structured responses and a more strategic (deep thinking) response.
Example prompt:
“Let’s walk through this problem together: I need to launch a new product with a limited budget and a small team to appeal to an already crowded market. What’s the best strategy?”
Use it when: You’re building timelines, weighing trade-offs, or need a coherent plan that holds up under scrutiny.
Copilot Researcher Agent: Deep Dives and Source-Backed Intelligence
Best for:
Comprehensive research and reports
Source comparison and opinioning
Insight generation
How it thinks: Researcher mode is a custom build agent from Microsoft specifically designed to work in a recursive, self evaluation and exhaustive approach. It pulls from multiple sources which it compares and contrats against one-another, compares methodologies, and identifies gaps. This is design for for thought leadership, academic-style or research style analysis, or strategic decision-making with different view points.
Example prompt:
“Find and summarise studies on the effects of remote work on productivity. Compare methodologies and results, looking at pro’s cons, impact on society, well-being, productivity and trends across different EU regions.”
Use it when: You need more than just an answer. You want to combine different view, present the otput in a particular way and when you need comprehenisive evidence, nuance, and a narrative that stands up to boardroom or peer review.
Which One When?
There is no definitiave answer to this, but based on my usage and infomration I have read in the field, guidance from OpenAI, Microsoft and others, this table provides a guide to which mode would work best in the scenaios below
Task Type
Use GPT-5 Chat
Use GPT-5 Reasoning
Use Researcher
Write a quick summary
✅
Build a strategic roadmap
✅
Compare academic studies
✅
Brainstorm creative ideas
✅
Solve a complex problem
✅
Generate source-backed insights
✅
AI isn’t a one-size-fits-all and not all use cases are the same. As AI models contine to become more context aware and with the ability to switch modes, there are still many times where knowing where to take your problem to is key – just like you know to talk to a specialist consultant vs a generalise at times.
Currently, to get the best from these AI models, the best result often lies in knowing when to switch gears—from conversational to analytical to research-grade depth. Whether you’re a CTO shaping strategy or a content creator chasing clarity, the right mode turns AI from assistant to partner to mentor.
Would you like this adapted into a SharePoint-friendly format or paired with a branded visual? I can also tailor it for internal enablement or customer-facing use.
Teams has a powerful new capability called the Facilitator Agent – a Copilot-driven meeting assistant designed to make collaboration smoother, smarter, and more productive. Think of it as a virtual chairperson that keeps your meeting on agenda, on-time and to point,whilst allowing participants to focus more on the meeting than taking notes.
Facilitator in Teams Rooms – Image (C) Microsoft.
Facilitator auto-drafts agendas, keeps people on track of the agenda and timings, provides rolling summaries, decisions, and action items all in a secure shared Loop page that everyone can co-author / edit across desktop, web, mobile, and even now in Teams Rooms direct from the room controls in Teams Rooms.
What is the Teams Facilitator Agent?
The Facilitator Agent is an AI-powered feature built into Microsoft Teams that works alongside Microsoft 365 Copilot (you need a Copilot license to activate it and interface to it). It acts as a shared assistant within your meetings and chats, providing:
Real-time AI-generated notes: Captures discussion points, decisions, and action items as the meeting unfolds.
Collaborative editing: All participants can view notes and Microsoft 365 Copilot licensed users can co-author notes live – this ensures accuracy and inclusivity.
Meeting moderation: Helps manage agendas, prompts for goals if none are set, and even nudges participants to wrap up discussions.
Time management: Includes a meeting clock and reminders to keep sessions on schedule.
Post-meeting recap: Provides a structured summary and tasks in the Recap tab, stored securely in Microsoft Loop in the meeting organisers’ OneDrive.
How is it Facilitator different from the old “AI Notes” feature?
Previously, Teams offered AI Notes as part of Intelligent Recap, which generated summaries after the meeting. While useful, it was a passive experience—participants couldn’t interact with or influence the notes in real time.
The Facilitator Agent replaces and enhances this by:
Working live during the meeting, not just after.
Real-time co-authoring of notes by both AI and humans as the meeting progresses.
Acting as an active participant, responding to @mentions and questions in chat.
Providing dynamic updates as discussions evolve, rather than static summaries.
Keeps a track of the meeting, who has spoken, actions and topic/agenda drift (in otherwords it politely nags you!)
What is Facilitator good at?
Facilitator can or could if trusted, replace the chair or act as a chair/co-chair in a meeting. In my personal experience I have foudn it to be really really good at:
Real-Time Note-Taking & Summarisation Capturing key discussion points, decisions, and action items during meetings, with live co-authoring – I love how it writes as the meeting prgresses and even corrects itself.
Meeting Moderation & Structure Detects if a meeting lacks an agenda and prompts participants to define goals. If a meeting has an agenda it attempts to chunk the meeting into sections and helps keep the meeting on topic and ontime.
Improved Collaboration Works in meetings and group chats, keeping everyone aligned—even late joiners. It allows people to talk to the agent too – by mentioning @facilitator if you need it to do something like set an action or recap a point.
Post-Meeting Recap & Accountability Generates structured summaries and suggested tasks in the Recap tab for people to go back to or generate an email follow from etc,.
Facilitator Agent – Current Limitations
I do love usig it – its been GA for a few weeks now, but there are few limitations which I hope/expect will “go away soon”. These include:
Not Available Everywhere: Facilitator currently doesn’t work in external, instant, or channel meetings; mobile users can view notes but not start Facilitator (yet)..
Compliance Gaps: Sensitivity labels don’t automatically apply to notes yet but this is in thge public roadmap.
Using Facilitator in Meetings
Turning it on: By Default, when you create a meeting via Teams, Facilitator is “off” and needs to be enabled by switching the toggle as illustrated below. It can also be enabled from “within” the meeting.
In Meeting Interaction: When the meeting starts, you are notified that Facilitator is running via an in-app notification. Note the meeting does not need to be recorded for this to be active. You also see this indicator under the notes section at the right of the meeting pane.
By the way, if you join a meeting where Facilitator is not active, you can enable it anytime from the menu under “…more”.
You still get a notification when Facilitator is running, and it will period chat to you in the meeting chat to keep you updated on the meeting.
Facilitator in Meeting
In Meeting – Meeting Notes and Actions Beng taken by Facilitator
Actions Generated by Facilitator
During the meeting (and afterwards, which you can find by going back to the meeting in your calendar), you can view and of course edit the notes, actions and also see any “related” content and “insights” that Facilitator has sufaced that it “thinks” might be relevant to the meeting dsicussion you have been in. These notes are captured in a Loop Space which is stored on the meeting organisers OneDrive and shared (automatically) with all meeting participants.
Post Meeting Notes, Actions and Insights.
Facilitator Agent Use Cases
I use this in most meetings but there are loads of use cases I see and hear about.
Daily stand-ups or project huddles to log progress and blockers
Customer calls and scoping meetings capturing commitments and next steps to eliminate follow-up churn
Project update or planning calls.
Facilitator Agent – What is Coming (Roadmap)
This is in preview and will be fully rolled out (GA) by September, and there are a few thinsg still nt he works which I expect will be out soon enough.
Editable Canvas for Chat Notes: AI notes in chats will move to an editable canvas backed by SharePoint Embedded.
Teams Rooms Integration: Facilitator will also (now in Preview) support ad-hoc and scheduled meetings in Teams Rooms, with QR code invites and speaker attribution.
Improved Compliance: Sensitivity label inheritance and enhanced governance via Microsoft Purview will be supported.
Q & A
Q: Is the Facilitator Agent just a rebrand of the previous AI notes feature? A: It builds on that toggle but expands into a full-blown agent. Beyond post-meeting summaries, Facilitator prompts agendas, generates live recaps, drives collaboration via Loop, and integrates with Teams Rooms by QR code.
Q: How does it differ from using Copilot in a Teams meeting? A: Copilot in a meeting is a private assistant—only you see its responses. Facilitator operates in the group context: prompts, highlights, and action items appear for everyone to view and edit in real time.
Q: What’s the added value over just recording and transcribing? A: Recording and transcription are passive: you consume them after the fact. Facilitator is proactive—drafting agendas, nudging for goals, surfacing decisions, and giving every attendee an editable canvas mid-meeting.
Q: Where does Intelligent Recap fit in? A: Intelligent Recap synthesizes speech and on-screen visuals after the meeting ends. Facilitator closes the loop instantly – keeping the conversation structured, accountable, and collaborative from start to finish.
Q: What are the alternatives to Facilitator Agent?
1. Native recording + transcription then manual or Copilot/ChatGPT note generation 2. Intelligent Recap for post-meeting slide and data context 3. Private Copilot chats for ad-hoc AI queries 4. Manual note-taking or shared OneNote pages 5. Third-party assistants like Otter.ai or Fireflies.ai
Q: Do I need a Copilot license to use the Facilitator Agent? A: Any user who initiates or edits AI-generated notes in meetings or chats must have a Microsoft 365 Copilot license. Unlicensed participants can view meeting AI notes but cannot start or edit them.
Q: What about in-person meetings? A: Coming soon – a new feature in the Teams mobile app will let you start a dedicated in-person meeting with Facilitator right from your phone. This will then kick off a recorded, transcribed session – again with real-time agendas, notes, and follow-up tasks. When you end the meeting, notes save automatically and a “in the past” calendar event is created—everything is surfaced in Recap. – This will requires a Copilot license and is due to be in preview Auguist/Sept – I’ve seen it but don’t have it yet myself!
We are in the middle of rapid shift – AI agents are no longer just reactive helpers waiting for a us to give them a prompt. Instead, they are becoming proactive, and autonomous , capable of initiating actions, orchestrating workflows, and making decisions across systems.
If you’ve already built governance models for low‑code platforms like Microsoft Power Platform, you’re not starting from zero. Those same principles – with a few smart extensions can help you govern the next generation of agents built in Copilot Studio.
What is Agent Governance? Agent governance encompasses the rules, policies, and oversight mechanisms that guide the behavior of AI agents – autonomous systems capable of performing tasks with minimal human intervention. This governance is crucial to ensure that these agents operate in a manner that is legally compliant, ethically responsible, and operationally safe!
Microsoft have shared new blue prints and guidance to help you get started with healthy goverance for Copilot Studio – which I have linked to and summarised below…
1. Lead with a Governance Mindset
Agents aren’t “just another app.” They’re digital labour – they (can) talk across systems and across roles and need managing just like humans. This means they they need:
Trackable identities — so you know exactly which agent did what, and when.
Scoped permissions — the principle of least privilege applies here too.
Continuous oversight — because autonomy without accountability is a risk.
Not every agent should have the same freedom. For example, a Q&A bot answering FAQs is low risk. An autonomous sales development agent drafting proposals is much higher stakes and an agent that takes a customer interaction and acts on it automonously is high risk.
We must define tiers of autonomy and enforce them with technical guardrails.
2. Apply Your Low‑Code Lessons
If you’ve governed Power Platform, you already have your own playbook:
Managed environments to separate dev, test, and production.
Role‑based access control (RBAC) to manage who can create, deploy, and run agents.
Data Loss Prevention (DLP) policies to control what data agents can access or share.
Audit logs to track behaviour and support compliance.
These aren’t “nice to haves” — they’re essential for safe, scalable agent adoption. Extend your existing frameworks to cover new agent behaviours.
3. Drive Visibility, Cost Control, and Business Value
Governance isn’t just about control — it’s about clarity. Visibility and telemetry is really important becuase it tells us:
Who created the agent.
What data it touches.
How often it’s used.
The business outcomes it’s driving.
With that visibility, you can spot redundantagents, forecast costs, and focus investment where it delivers the most value. Tools like Copilot Studio analytics and Power Platform Admin Center make this possible — but only if you use them consistently.
4. Empower Innovation with Guardrails
The people closest to the work often have the best ideas for agents. Advice is to empower them to experiment — but within a zoned governance model:
Zone 1: Personal Productivity — safe sandboxes for individual experimentation.
Zone 2: Collaboration — team‑level development with stronger controls.
Zone 3: Enterprise Managed — production‑grade agents with full monitoring and lifecycle management.
This approach balances speed and safety, enabling innovation without compromising compliance.
5. Build Community, Training, and Experimentation into the Culture
Governance is as much cultural as it is technical and it’s the culteral and human aspects that typically impact and slow adoption.
A thriving Center of Excellence (CoE) should:
Host “Agent Show‑and‑Tell” sessions and hackathons.
Appoint champions in each department to mentor others.
Provide role‑based training for makers, admins, and business leaders.
Encourage responsible experimentation — and celebrate successes.
As with any transformational shift, when people feel supported and inspired and part of the journey, adoption accelerates and impact flourishes.
Why This Matters Now
According to Microsoft, over 230,000 organisations – including 90% of the Fortune 500, are already using Copilot Studio, and IDC projects there will be a staggering 1.3 billion AI agents by 2028.
This scale and exponential speed of adoption make governance a critical priority, not an afterthought or option!
The CIO’s role is shifting from enabling agents to governing them at scale — ensuring they’re secure, compliant, cost‑effective, and aligned with business goals. That’s not just a technical challenge; it’s a leadership opportunity.
Summary – the Key Steps
Extend your low‑code governance — apply your Power Platform controls to agents.
Define autonomy tiers — match oversight to risk.
Instrument for visibility — track usage, cost, and impact.
Microsoft has introduced the =COPILOT() function in Excel, embedding AI directly into spreadsheet cells. This formula turns natural-language prompts into structured outputs—no VBA, no complex formulas—so anyone can perform advanced analysis with a simple cell entry.
This essentially turns your prompts into Excel formulas direct from your excel cell!
Copilot() function in excel
This is in currently in public preview.
What It Is and How It Works
The COPILOT() function behaves like any native Excel formula. You type =COPILOT("andyourprompt", [range]) in a cell, and Excel sends the request to the Copilot service (powered by Bing and ChatGPT). The AI returns a grid-friendly result that recalculates automatically whenever your source data changes. You can even nest COPILOT() inside functions like IF or LAMBDA for more sophisticated logic.
Core Capabilities
Summarise, group, or categorise data using plain-English prompts
Perform sentiment analysis on text feedback
Extract and organise information from unstructured text (names, emails, URLs)
Generate dynamic lists, schedules, or qualitative ratings
Augment tables with symbols or simple markers for clearer storytelling
Key Use Cases
Automating data cleanup: standardise formats, remove duplicates, split columns.
Customer insights: turn free-text reviews into sentiment scores and themes.
Sort data and represent in different formats without having to learn how to create pivot tables.
Transforming data using formulas without having to write a formula.. Just natural language.
Prerequisites & Access
To use COPILOT() in Excel, you must meet the following requirements:
Microsoft 365 commercial Copilot license (not included in Personal/Family plans)
Microsoft Entra ID account and primary mailbox on Exchange Online.
Excel Beta Channel build 19212.20000 or later / macOS build 25081334 or later
Up to 100 function calls per 10 minutes (300 per hour); use array inputs to conserve quote.
Data stored in the active workbook (external sources not yet supported
How to access:
In Excel, go to File > Account > Office Insider and switch to the Beta Channel (Windows).
On Mac, open Help > Check for Updates in Microsoft AutoUpdate and choose the Beta release.
Sign in with your work/school account that has a Copilot license; use File > Account > Update License if needed
Restart Excel—=COPILOT() will now be available in any cell, or via the Home > Copilot pane.
Requirements and Limitations
Not optimised for heavy numeric or matrix computations
Outputs are dynamic—save critical results as values to prevent unintended changes
Only works with in-workbook data; live web or external data access is pending
Why It Matters
Excel remains the lingua franca of business data. By transforming the grid into an interactive AI canvas, COPILOT() tears down formula-syntax barriers, accelerates decision-making, and empowers every user—from analysts to frontline managers—to become data storytellers. Enablement leaders can shift focus from formula training to writing effective AI prompts and compelling narratives.
In short, it’s powerful for people that are not excel formula wizards!
Yes… Microsoft are updating Microsoft 365 Copilot with support for GPT-5 across the Microsoft Al stack. This is live now and rolling out across Microsoft 365 Copilot and Copilot Studio after being made available to “Insiders” on Copilot consumer/personal last week.
This quick incorporation of GPT-5 into Copilot underscores Microsoft’s pledge to integrate OpenAI’s cutting-edge models into the their AI products within 30 days of availability.
What is so great about GPT-5?
GPT-5 in Copilot is built on a dual-engine (think two brain approach) architecture designed to better align to the way humans think. It will. Adapt the “mode” based on the ask and type of response of work needed.
Real-time routing: rather having to choose the model (such as deep thinker or research), for every prompt, Copilot now automatically evaluates the prompt and it’s complexity and then selects the ideal GPT-5 sub-model for the response.
High-throughput model : Tackles routine tasks quickly, delivering succinct answers to straightforward requests.
Deep-reasoning model: Which engages when advanced analysis or creativity is needed, taking time to plan, verify context, and ensure accuracy before responding.
This adaptive model selection brings together speed and depth, and can change within the same conversation. This means as your conversation with Copilot evolves so does the way it responds, without the user having to change modes.
(c) Microsoft
Open AI’s CEO Sam Altman said that “the new model, GPT-5, is its smartest and fastest to date with wide-ranging improvements to ChatGPT’s skills in areas like coding, writing and taking on complex actions.”
How GPT-5 in Copilot shifts the conversation.
GPT-5 builds on GPT-4 with:
A vastly expanded context window (up to 100K tokens) – on average a token is equivalent to about four characters of an English word.
Improved reasoning and multi-step problem solving without having to manually choose the model up front. Can also switch dynamically in the same conversation.
Enhanced memory and recall capabilities
Support for multimodal inputs (text, image, audio)… Note output is still text!
Faster, and much more accurate responses
Copilot with GPT-5 isn’t just smarter—it’s more practical too. The increased token input also means it can handle entire project folders, analyse longer documents, and deliver context-aware outputs that feel tailored to your workflow and chosen format.
Image a scenario where I ask Copilot to “Summarise a new solution proposition“. This would trigger GPT-5’s high-throughput route, scanning the document or documents to return a concise summary as per the ask.
When I then ask Copilot to “review this against best practise examples, and make suggestions fornimprovement and then create me a ‘better’ version based on your suggestions“, Copilot will seamlessly switch to use it’s deep reasoning mode. You know it has switched modes as you literally see it think.
GPT-5 Agents in Copilot Studio
People or teams building specialised workflows or agents in Copilot Studio now also get support for GPT-5 which is now the primary engine for custom agents.
GPT-5 better enables agents to tackle more complex processes like compliance audits or financial modeling with greater precision and contextual awareness than previous GPT powered agents.
How to use GPT-5 in Microsoft 365 Copilot
GPT-5 support in Copilot is available now for licensed Microsoft 365 Copilot users. Once rolled out to your environment, users will see a new “Try GPT-5 button” in Copilot Chat. Once activated, Copilot will leverage GPT-5 across your work and web data by default.
A Microsoft 365 Copilot License grants priority access, with broader rollout to Copilot Chat only users rolling out over the next few weeks.
Is it better?
My suggestions
Compare a prompt such as: “Summarise my emails from the last week, determine which ones require actions and break them down into high impact, low impact and trivial based on your analysis“.
Try this and try this again with GPT-5 enabled.
Try longer prompts…you can essentially now. Feed Copilot with…
A full business proposal
A multi-tab Excel workbook
A folder of Markdown files or code
A long-form research paper with citations
…it can handle all of it in one prompt, as long as the total token count stays under 100K.
It’s important to note that the prompt is just part of the context window. The model also needs room for its response. So if you use 80,000 tokens for input, you’ll have ~20,000 tokens left for output. Hopefully that makes sense.!
This isn’t just summarising anymore—it’s deep analysis, synthesis, and contextual understanding across dozens (or hundreds) of pages.
It snuck in quietly, like all meaningful innovations do. I didn’t see a press release, or announcement – and just saw it “pop” up for me today with one small pop-up. Just one single word, Smart. And yet, beneath the understated label lies perhaps the most pivotal shift in the way generative AI models like Copilot and ChatGPT work since manual “model selectors” first became a thing.
This is, yes, you’ve guessed it GPT5!
Smarter Than Smart – the quiet revolution of reasoning
Copilot (i’m talking the consumerversion currently at copilot.microsoft.com) or via the Windows, IOS and Android app, currently has three modes of chat which you choose based on the discussion with Copilot you want to have. This is similar to how ChatGPT works also today.
Quick Response [for every day conversations]
Think Deeper [for more complex topics]
Deep Research [Detailed reports with references]
That is changing – Microsoft Copilot’s new ‘Smart’ mode doesn’t ask you to choose a conversation type anymore. Instead it now adapts for you – automatically and intuitively.
This means, that depending on your query, for example whether you’re scoping customer insights, untangling a tricky dependency in a network diagram, or storytelling your way through a general chat, summarisation of marketing ideas, ‘Smart’ mode calibrates itself to the conversation and task at hand.
In short – in this mode, Copilot will now decide what model it thinks it needs to help you. Copilot has auto reasoning — true adaptive, context-aware reasoning based on the ask.
What Makes This Mode Smart?
‘Smart’ mode is likely powered by OpenAI’s upcoming GPT-5, a model anticipated to merge the OpenAI o-series and GPT-series models into one unified framework.
What we’re seeing as these models evolve is:
An intuitive reasoning engine, not just predictive text- task “and” context aware
Self-calibrating depth, reducing cognitive load by using the right tool for the right job
Model abstraction, freeing users from having to pick the right tool for the job themselves
Microsoft hasn’t just added another dial. It’s looking to hide the dial entirely — and taught the model how to turn it for you (however as it’s in preview you do need to turn the auto mode on – for now at least).
Copilot’s Human Centric UX
As you can see above, instead of users needing to flip between “Think Deeper,” “Quick Response,” and “Deep Research,” (or not evening understanding what these mean and therefore ignoring it), Copilot’s Smart mode does what most tools never do: it assumes responsibility. That’s more than a UX shift — it’s a culture shift. This means no longer asking (non technical) users to understand what the different models the model hierarchy or decoding acronyms like o4-mini. Instead Copilot is getting cognitive delegation.
This means we will be able to “trust”Copilot to know when to dig deeper and when to skim the surface.
Examples: Tech Architects, Storytellers, and Strategists
The table below gives some examples of where Copilot Smart mode can make a big different in use:
Role
Before Smart Mode
With Smart Mode
Solution Architect
Manually toggling depth based on task complexity
Instant adjustment to scope and context
Content Creator/Marketing
Selecting modes based on tone and detail required
Natural flow from quip to deep dive
Enginner
Testing prompts for clarity vs depth
Getting both — with nuance — the first time
This isn’t just about being faster. It’s about being right-sized. Strategically aligned, creatively agile, and cognitively respectful.
What about control and “mode anxiety”?
We’ve all wrestled with prompt engineering, hoping we’re not asking too little or too much. Smart mode is Copilot whispering, “I’ve got you.” That’s a leap from assistant to partner — the kind we’ve spent decades trying to design into our workflows, team cultures, and tech stacks.
Copilot Smart Mode Preview?
This is in preview clearly – or was it rolled out silently. Anyway, if you have it, give it a try (I have it on desktop and web plus mobile). Let me know your thoughts.
It will be interesting to see if eventually the modes disappear and we just have an “auto” mode.
Copilot Memory is a new capability within Microsoft 365 Copilot (similar to what ChatGPT has) that allows Copilot to remember key facts about your preferences, working style, ongoing projects, and other things you want it to know about you. This enables it (think PA) to be able to tailor its responses over time. You can add and change this as needed so it evolves with you, reducing repetitive prompts, adapting to your style and speeding up your daily tasks.
Key Capabilities
Persistent Facts
Copilot picks up on explicit instructions like “Remember I prefer bullet points in my writing” or “Always use a formal tone in emails” and retains these details across sessions.
Custom Instructions
Beyond passive memory, you can proactively shape Copilot’s baseline behavior. Ask for brevity, wit, or a specific document style, and Copilot applies those instructions automatically in Word, Excel, Outlook, and other 365 apps.
Contextual Recall
Copilot integrates with Microsoft Graph and ContextIQ to ground conversations in your files, meetings, and chats, ensuring its outputs align with your latest work context.
How It Works
Explicit Memory Prompts
Copilot only stores information when you ask it to. This prevents unwarranted data collection and keeps your AI focused on what matters to you.
Memory Updated Signal
Whenever it logs a new fact, you’ll see a subtle “Memory updated” badge—confirmation that Copilot has learned something new about your preferences.
Privacy Controls
You can control its memory: You can view, edit, or delete entries in Copilot’s Settings pane and if you need to can wipe it’s memory and start fresh by simply toggle the Memory function off entirely.
Admin and Compliance Oversight
Organisations can disable Memory for specific users or tenant-wide, and all memory actions flow into Purview eDiscovery for audit and compliance purposes.
Timeline & Availability
Rollout date: July 2025 (staged)
Why Copilot Memory Matters
Efficiency Gains
This is really about efficiency and personalisation since you will no longer need to keep telling Copilot your preferred tone or formatting preferences. This speeds up document creation, email drafting, and data analysis.
Deep Personalisation
By remembering your recurring topics—Project Alpha, Python for data science, or icon-size images—Copilot provides responses that are more tailored to each user, not generic AI outputs.
Enhanced Adoption
For organisations, personalised AI interactions drive higher engagement and adoption of Copilot across teams, leading to greater ROI on AI investments.
Trust & Transparency
Visible memory updates and clear controls build user confidence in the AI, ensuring you always know what Copilot retains and why.
Enabling Copilot Memory
Memory is an option feature and can be enabled, modified and disabled as needed. To enable it, follow the instructions below.
Open Microsoft 365 Copilot and head to Settings › Account › Privacy.
Under Personalisation & memory, toggle Memory on or off.
Tell Copilot what to remember: “Remember I prefer bulleted lists,” or “Keep my summaries under 100 words.”
View, edit, or delete memories any time from the same settings pane.
So, I was delivering an AI awareness session earlier this week to 120 eight- to ten-year-olds at my sons’ school. In the Q&A bit, one of the children asked me – “What is Agentic AI?”. Needless to say, I was quite surprised to be asked that question by a nine-year-old. We actually decided to us AI to explain it to the class…. which I must say it did a better job that I would have done.
Agentic AI Explanation for a ten-year-old.
Here is what AI said: “imagine you’ve got a super-duper smart robot friend. This friend doesn’t just follow orders and obey instructions like a regular robot. It is like a super robot, the Dyson of Robots. This Robot can make decisions on its own, like deciding what to do next or how to do something when you haven’t taught it. ‘
Agentic AI’ is like that robot friend, but way more advanced. It can make its own choices, set its own goals, and figure out how to achieve them, almost like it’s got its own mind but by following a set of guidelines and rules. So, it’s not just reacting to commands, it’s thinking ahead and planning like a little strategist!”
Agentic AI for everyone else
I was then thinking, well what if you don’t work in an IT role. Does everyone know what Agentic AI is? It is certainly the buzz term of 2025 and certainly a leap forward from the “chat bots” we have played around with over the past 2 years or so with ChatGPT, Copilot etc.
I was then thinking about how I need to explain Agentic AI to some my co-workers so in my usual style I wanted to write it down to get my wording right. here’s what I came up with which I think and hope anyone can understand. As such I wanted to share it.
So…..
Think of agentic AI as more of “system” than a chat bot. Unlike a chat bot which is generally more about responding to a request or returning information, Agentic AI operates with a high degree of autonomy. Rather than just follows predefined instructions or responding based on information it has been fed/trained on, agentic AI can set its own objectives and determine. by itself, the best course of action to achieve them. This is a very different approach to what we have seen before now since it can not only executes tasks but also identifies opportunities, develops strategies, and takes initiative without constant oversight or being asked.
This has the potential to be a powerful tool in many different roles and organisations. Here’s a few examples I have pulled together based on some of the customer converations and usecases we are exploring at the moment.
Agentic AI Use Cases
Healthcare : Agentic AI could proactively identify potential health risks in patient data, following or before treatment, suggest personalised treatment plans, and even coordinate with pharmacy and supply chains to ensure medication availability. It could even be used to help patients better understand their health and nurses better explain to patients.
Gym: It could create personalised workout plans for members, monitor equipment usage to predict maintenance needs, and even suggest new classes based on emerging fitness trends. For Mangement it could suggest changes to class schedules based on enquiries, booking history, attendance etc.
Retail : It could autonomously manage inventory, predict trends by analysing customer data, external factors such as weather, news events etc, and even optimise pricing strategies based on market demand and competitor analysis such as changing the price of suncream when it gets hot and the price of umbrellas when it rains.
Public Sector : It could streamline citizen services, anticipate infrastructure needs based on usage patterns, and improve disaster response by dynamically allocating resources. It could also pre-empt and influence bin collections based on realtime data, or take proactive action and make recommendations from transcripts based on interviews or care notes in social services.
Legal: It could autonomously manage case documentation, chase up cases, predict case outcomes based on historical data, and even recommend legal strategies or layers most likely to win particualr cases. It could provide guiance to customers, based on “learned” cases for that firm and provide “virtual lawyer” services fully automonosly.
Insurance : Agentic AI could assess risk profiles, help detect fraudulent claims, and tailor policy recommendations to individual customers.
School Admissions : It could predict enrollment trends, identify potential gaps in student demographics, and optimise the selection process to ensure a diverse and well-balanced student body.
These are just a few examples of Agentic AI’s ability to act independently and adapt to complex, changing environments makes the applications and use cases almost endless as long as we can guide it, trust it and step in when needed.
Microsoft 365 Copilot now includes two advanced AI agents – Researcher and Analyst that became generally available in this month ( June 2025). These agents use powerful reasoning models (based on OpenAI’s o3-mini and deep research models) to handle complex tasks beyond what the standard Copilot could do.
Researcher is a specialised agent for multi-stepresearch – it can securely comb through your work data (emails, files, meetings, etc.) and the web to gather information, ask clarifying questions, and produce well-structured summaries and insights. It’s ideal for tasks like market research, competitor analysis, or preparing for big meetings – work that used to take hours, now done in minutes with higher accuracy.
Analyst is a virtual data analyst/data scientist built into Copilot. It excels at advanced dataanalysis, working through messy spreadsheets or databases step-by-step using chain-of-thought reasoning and even running Python code when needed. From identifying sales trends to spotting anomalies in finance data, Analyst gives you in-depth answers and visuals that mirror human analytical thinking.
Compared to the standard Microsoft 365 Copilot, these agents go much further in reasoning and capabilities for these specific tasks. While the native Copilot mod helps draft documents or summarise content, Researcher and Analyst tackle complex reasoning tasks (deep research and data analysis) with a level of thoroughness and skill akin to an expert – essentially “like having a dedicated employee at your side ready to go, 24‑7,” according to Microsoft’s Jared Spataro. They are accessed through the Copilot interface (pinned in the Copilot app and via Copilot Chat) and come with a usage limit of 25 queries permonth per user due to their intensive workloads.
Analyst vs. Copilot for Finance:
Analyst is a general-purpose data analysis agent available to any Copilot user, whereas Microsoft 365 Copilot for Finance is a separate, role-based Copilot designed specifically for finance teams. Copilot for Finance connects to financial systems (like Dynamics 365 and SAP) and Microsoft 365 apps (Excel, Outlook) to automate finance workflows (reports, reconciliations, insights). Unlike the Analyst agent which works on data you provide, Copilot for Finance directly taps into live enterprise finance data for real-time insights. Importantly, Copilot for Finance is not limited to Dynamics 365 – it can integrate with various ERPs including Dynamics 365, SAP, etc via connectors though it is deeply optimized for Dynamics 365 Finance.
The Age of AI Specialists in Microsoft 365 Copilot
Microsoft 365 Copilot is evolving from a single assistant into a team of AI specialists. Earlier this year, Microsoft announced two first-of-their-kind “reasoning agents” for work: Researcher and Analyst. After a period in preview (through the Frontier program) for early adopters, these agents are now generally available to all users with a Microsoft 365 Copilot license as of June 2025. This marks a significant expansion of Copilot’s capabilities beyond its initial skill set.
The new Researcher and Analyst are advanced Copilot modes (agents) specialised for particular scenarios – complex research and data analysis. They join other Wave 2 Copilot features (like the new Agent Store, Copilot Search, Memory, Notebooks, and image generation) that Microsoft has been rolling out to enhance the Copilot experience. Jared Spataro, Microsoft’s CMO for AI at Work, describes these agents as delivering “advanced reasoning” and notes “it really is like having a dedicated employee at your side ready to go, 24-7.” In other words, Microsoft 365 Copilot is no longer just a helpful assistant within Office apps – it can now also act as an on-demand subject matter expert that tackles higher-order tasks.
From a technology standpoint, both agents leverage the latest AI models tailored for their specific domains. They use OpenAI’s powerful models (codenamed o3-mini for Analyst, and a deep research model for Researcher) combined with Microsoft’s orchestration, search, Responsible AI, and tool integrations. This means they don’t just generate quick answers; they actually reason through problems in multiple steps, consult various data sources, and produce more comprehensive results. This blog explores each agent in detail:
Microsoft 365 Researcher Agent
Researcher is the new Copilot agent that acts as a highly skilled research assistant. It’s designed to help you tackle complex, multi-step research projects right from your Microsoft 365 environment. Researcher brings together OpenAI’s “deep research model” with Microsoft 365 Copilot’s advanced orchestration and search. In practice, this means it can scour both your organisational data *and* external sources on the web to find the information you need, synthesize it, and present insights in a coherent way.
What can Microsoft 365 Researcher Agent do?
Microsoft describes Researcher as “an agent that can analyse vast amounts of information with secure, compliant access to your work data – your emails, meetings, files, chats, and more – and the web” to deliver expert insights on demand. In simpler terms, Researcher is great at doing all the digging for information, reading it and then summarising the findings for you. Some of its capabilities include:
Multisource Information Gathering: It can search through your files, emails, SharePoint, and external online / Web sources to collect relevant data and. For example, if you’re exploring a new market or analysing a topic, Researcher will pull from both internal documents and credible websites to gather material.
Smart Summaries: After collecting information, Researcher summarises what it finds in plain, easy-to-read language. You get a clear, tailored report instead of a dump of raw data. It will highlight key points, trends, and insights rather than making you sift through hundreds of pages or search results.
Trend and Insight Identification: Researcher uses its AI reasoning to spot patterns, trends, and opportunities in the information. It can draw connections and highlight things that might make a difference for your project or question. For instance, it might notice an emerging customer preference across feedback data or identify a common thread in market research reports.
Interactive Refinement: If your initial query is broad, Researcher often asks clarifying questions to narrow down the scope and ensure it’s on the right track. This interactive back-and-forth helps it deliver more relevant results. You can guide it by answering those questions or giving additional instructions, much like you would with a human researcher.
Citations and Source Transparency: When delivering its findings, Researcher provides well-sourced content. It can include citations or references for where information came from, so you can trust but verify the results. (This is crucial for workplace research, and you can ask it to only use authoritative sources for extra confidence, as in one example prompt Microsoft shared).
Use Cases for Microsoft 365 Researcher Agent
Researcher is great in situations where you need to quickly learn or compile knowledge on a topic or subject area but are not sure where to look. This could be for tasks like assessing the impact of the new Trump tariffs on business lines, preparing for vendor negotiations by gathering supplier intel, and collecting client research before sales pitches.
Researcher Agent Example
In a business context, imagine your sales / marketing team are looking for a fresh perspective on top technology investments organisations are making in the UK based on industry research which needs to be in a report. You could ask Researcher “What are the top technology investments and projects by “small to medium” and enterprise organisations in the UK. Use trusted market data from repuatble sources such as Gartner, IDC, Cisco, Microsoft, Canlays, CRN etc.”
What I love is how you see the deep thinking and reasoning Researcher is using to compile the information and generate your report. This is so much easier than manually searching the web and reading dozens of articles. Instead, Researcher gives you a report in just a few minutes.
Instead of manually having to search the web and read loads and loads of articles, Researcher gives you a report in under ten minutes. You can of course tweak the response by asking more questions or requesting adjustments to ensure it meets you needs. When the report is finished you’ll see how comprehensive and well formatted it is, allowing you to export to, add it to a collaborative Copilot Notebook or leave it as is.
Sample output from Researcher Agent.
Microsoft 365 Analyst Agent – Data Analyst
If Researcher is your content and knowledge scout, Analyst is your number-crunching, data-savvy AI team member. The Analyst agent is all about diving into data (often numerical or structured data) to extract insights, find patterns, and answer complex analytical questions. Microsoft describes Analyst as “thinking like askilled data scientist”, using an advanced reasoning approach to tackle data problems step-by-step https://www.microsoft.com/en-us/microsoft-365/blog/2025/06/02/researcher-and-analyst-are-now-generally-available-in-microsoft-365-copilot
What makes Microsoft 365 Copilot Analyst Agent special?
The Analyst agent runs on a finely-tuned AI reasoning model (post-trained on OpenAI’s o3-mini model specifically for analytical tasks). Unlike a standard chatbot that might try to answer a data question in one go (and often make mistakes), the Analyst agent uses a chain-of-thought process to break problems down and solve them iteratively. It can even generate and execute actual code (like Python) in the background to manipulate data, perform calculations, or generate charts. Throughout this, it adjusts to new complexities and can recover from errors autonomously – essentially debugging and refining its approach as it goes, much like a human analyst would. The end result is a thorough analysis with reasoning that is transparent to the user.
Here are some of the key capabilities of the Analyst agent:
Data Analysis Across Formats: Analyst can work with Excel spreadsheets, CSV/TSV files, databases, Power BI reports, and other structured datasources . It can even extract financial data from PDFs. It is possible to upload or point it to a dataset, even if the data is messy or hidden across multiple files. For example, if you have sales data split across a few different Excel sheets and files, you can use Analyst Agent to ingest them all. The agent can also clean up many of the typical issues found in spreadsheets such as wrong delimiters in a CSV, or values buried in an unexpected place before it starts to work. This means that your data does not need to be perfectly prepared beforehand .
Iterative Reasoning and Problem Solving: When you ask Analyst a question, it will hypothesise, test, and refine repeatedly. For instance, you might ask, “What insights can you find about our Q4 sales data, and why did some teams underperform?”. Here, Analyst might break this down into steps: first identifying overall sales by region, then noticing why one sales team is lower, then digging into possible factors (maybe inventory issues or lower marketing spend), then correlating that with other data. It takes as many steps as needed to arrive at a sound answer. This multi-step approach leads to more accurate and nuanced results than a one-shot response.
Code Generation and Execution: A standout feature – Analyst can write and run Python code behind the scenes to perform calculations or data transformations. If your data question requires a formula, statistical analysis, or creating a chart, Analyst will generate the code to do it. Even better, it shows you the code in real time as it works, so you have complete transparency into how it’s reaching its conclusion. You effectively have an AI that can program on the fly to solve your data problem. This is like having a data analyst who is also a programmer working for you instantly.
Insight Generation and Visualisation: Analyst doesn’t just provide text based results – it will also explain the “story” behind the numbers in plain language and can also create simple charts or graphs to illustrate key points. It could, for example, produce a trend line graph of sales over time or a bar chart of top-performing products if those help answer your question. It will highlight findings such as “Sales Team A had a 20% increase in Q4, outpacing their previous year results ,,,, ” By narrating and illustrating the data, it helps you quickly understand the business implications.
Actionable Recommendations: Analyst can often suggest next steps or recommendations based on the data patterns it finds. If it discovers, say, that a certain region’s sales are lagging due to low inventory, it might recommend increasing stock or marketing in that region. Or if a customer segment is showing poor engagement, it could suggest targeted outreach. These suggestions turn raw analysis into useful advice, bridging the gap from insight to action.
Microsoft 365 Analyst Agent Use Cases:
The Analyst agent is useful anywhere you have data and questions about that data. Some real-world examples Microsoft has noted include using Analyst to assess how different discount levels affected customer purchasing behavior to identify the top customers who aren’t fully utilising the products they bought, and to visualise product usage trends and customer sentiment for informing go-to-market.
Analyst Agent Example
In the example below, I took some Customer Support Tickets from an excel (see below).
Sample Customer Support Ticket Export
I then have asked the Analyst Agent to “review the support ticket and create me an exective summary of the tickets, pulling out trends and themes that my team should look at and how they might reduce future support call duration.”
The results below are the first run with data that represeted as I have asked.
How Do Researcher and Analysts Agents Compare to the Standard Microsoft 365 Copilot Experience?
With all the excitement around Researcher and Analyst, you might wonder how they differ from the core Microsoft 365 Copilot Chat experience that users have been trying out in apps like Word, Excel, Teams, and Outlook.
The key difference comes down to depth of reasoning and specialisation. The core Copilot Chat experience is like a well-rounded generalist – great at everyday productivity tasks, such as drafting an email, summarising a document or thread, writing in Word, generating a PowerPoint outline, or pulling insights from a single Excel worksheet. It uses a large language model (LLM) to understand your prompt and the context from the active document, then provides a response.
However, it typically gives a direct answer or action based on available content, without doing prolonged multi-step reasoning. For example, standard Copilot can summarise a document or create a draft from prompts, but if you ask it to perform a very complex analysis that requires digging through multiple files or doing calculations, it may hit its limits. Thats where these specialist agents differ:
Advanced Reasoning vs. Quick Responses: “Standard” Copilot Chat is designed for quick assistance within the flow of work (one-shot answers or short tasks). In contrast, Researcher and Analyst use advanced reasoning algorithms (chain-of-thought) that allow them to work through a problem in multiple steps). They will plan, execute sub-tasks (like searching sources and creating and executing code), and then refining its output. This means they can handle questions or tasks that the regular Copilot would either answer superficially or not manage at all.
Tool Use and Data Access: These specialist agents have access to a much broader set of information and models. Researcher can tap into web search and internal knowledge bases simultaneously, something standard Copilot doesn’t proactively do by itself. Analyst can use the equivalent of a built-in scripting engine (Python) to manipulate data. These abilities let the agents produce more accurate, data-backed results (for instance, Analyst can compute exact figures or generate a pivot table behind the scenes, rather than guessing).
Use Case Focus: Out of the box, Microsoft 365 Copilot has a breadth of capabilities across Word, Excel, PowerPoint, Outlook, Teams, etc., but each in a somewhat scoped way – e.g. helping write, summarise, or create within that app. It is “broad but shallow”. Researcher and Analyst are narrower but much deeper in their domains. If you don’t need multi-step research or advanced data analysis, you might not need to use them and the regular Microsoft 365 Copilot Chat or in app Copilot experience might suffice. But if you do have those needs, these agents provide a level of expertise that feels like a specialist joining your team.
For example, consider interpreting a complex financial report: Standard Copilot in Excel can summarise that report or maybe answer something about it if asked directly, but Analyst could take multiple financial files (ledgers, budgets, forecasts) and do a cross-file analysis, then produce a summary and suggest optimisations – a far more sophisticated outcome.
Interaction Model:Using Researcher/Analyst is a bit like launching a specific mode of Copilot meant for heavy tasks. They’re accessible via the Copilot app’s Agent Store or as pinned which is a different entry point than simply typing to Copilot in Word. This interface guides the user to ask bigger questions (“Help me investigate X” or “Analyse Y data for Z”) rather than the smaller in-app prompts. The agents also tend to show their working process (especially Analyst showing its code or reasoning steps), whereas standard Copilot just delivers the end answer in a friendly tone. This transparency is great for users who want to trust the results – you can literally see how Analyst arrived at an answer, step by step.
Analyst vs. Copilot for Finance – What’s the Difference?
With the introduction of the Analyst agent, you might also hear about Microsoft 365 Copilot for Finance – another AI offering that targets data and analytics, but specifically for finance professionals. It’s important to clarify how the Analyst agent and Copilot for Finance differ, because their names might seem related. In fact, they serve different needs:
Microsoft 365 Copilot for Finance (formerly introduced simply as “Copilot for Finance”, now in preview) is a role-based Copilot experience tailored for finance departments. This was announced in early 2024 as a way to “transform modern finance” by bringing generative AI into the daily workflows of finance teams. Unlike the Analyst agent – which any user with Copilot can use for various kinds of data analysis – Copilot for Finance is a separate add-on Copilot designed to integrate deeply with financial systems and processes. It essentially combines Microsoft 365 Copilot with a specialized finance agent and connectors to your financial data.
From what I have managed to assess these are the main differences between the Analyst agent and Microsoft 365 Copilot for Finance:
Aspect
Analyst Agent (Microsoft 365 Copilot )
Microsoft 365 Copilot for Finance
Purpose & Domain
General-purpose data analysis for any domain or department. Helps users analyse spreadsheets, databases, or other data to get insights.
Designed to work across certified and connected systems such as Microsoft 365 Dynamics, Salesforce and some others
Integration and Data
Works on provided or accessible data in Microsoft 365 (e.g. Excel files, CSVs, SharePoint data). No built-in direct connection to ERP systems – user typically uploads data or points to files for analysis
Connected to enterprise financial systems and data sources. Draws context from ERP systems (like D365 Finance & SAP) and the Microsoft Graph . Integrates in real-time with live finance data, assuming connectors are set up. Optimised for D365 Finance (seamless data access). Can connect other systems via custom or pre-built connectors).
Features and Skills
Uses chain-of-thought AI reasoning and Python code execution to perform analytics. Ideal for ad-hoc data analysis: e.g. combining sales data with customer data to find trends, identifying anomalies in operational data, generating charts from raw data. Acts as AI data analyst for any project.
Uses AI to streamline finance-specific processes and provide insights within finance workflows. For example, can automate variance analysis in Excel, perform reconciliations between systems, generate reports, summaries, and even draft emails for collections with relevant account info. Understands accounting principles and the company’s financial data.
User Experience
Accessed through the Copilot app as one of the agents (no special deployment beyond having Microsoft 365 Copilot license). The user asks questions or tasks in natural language and often provides the data files to analyze. The output is an interactive analysis in Copilot chat with optional visuals and code transparency.
Integrated into the tools finance teams use: primarily Excel, Outlook, and Teams in the context of finance work. For example, in Excel a finance user might invoke Copilot for Finance to run a budget vs. actual report or find anomalies in ledger data. In Outlook, it can summarise a customer’s account status from ERP data to help a collections officer. Works in flow of existing finance tasks, bringing AI where needed.
Availability & Pricing
Included as part of the Microsoft 365 Copilot (the Analyst agent is available to any user who has Copilot enabled). General Availability as of mid-2025. Usage is capped at 25 queries/month for heavy reasoning tasks.
Available as add-on to Copilot targeted at enterprises. Paid offering for organisations that use Microsoft 365 and want AI assistance in finance for supported systems like D365.
Dependencies on Microsoft Dynamics
Not dependent on Dynamics 365 – Analyst can analyse any data you give it. If your financial data is in Excel exports from SAP or Oracle, Analyst can still work with those exports, but it won’t directly pull from those systems on its own.
Deeply integrates with D365 Finance & Operations. Designed to plug into D365 modules so can act within that ecosystem (e.g., directly reading transaction data, posting results back). Through “connectors”, it can interface with other ERP or CRM systems too. Advantage is native use with D365 – without manual data exporting or integrations
To put it simply, the Analyst agent is like an AI data expert you can use for virtually any type of analysis by feeding it data, whereas Copilot for Finance is a comprehensive AI-powered solution built into Microsoft’s ecosystem to assist with a company’s financial operations in real-time. They might overlap in the sense that both can do things like variance analysis or finding trends in financial figures, but the context is different: Analyst would do it when you ask and give it the data (say, a couple of Excel files containing financial info), while Copilot for Finance would do it as part of your normal finance workflow, already knowing where the data is (in your ERP and Excel models) and proactively helping you in that domain.
Does Copilot for Finance only work with Dynamics 365?
No. Copilot for Finance is not limited to Dynamics 365, though that’s a primary integration. It brings together Microsoft 365 Copilot with a finance-focused agent that connects to your existing financial data sources including ERP systems like Dynamics 365 and SAP. So if your company runs SAP for finance, Copilot for Finance can use that data as well. Microsoft has built it to be flexible via connectors, because they know not everyone is on Dynamics. That said, organizations using Dynamics 365 Finance get a more seamless experience – Copilot for Finance can sit right inside the D365 Finance interface and offer insights without any data transfer.
In summary, Copilot for Finance is cross-platform in terms of data sources, but tightly integrated with Microsoft’s own finance solutions for maximum benefit. It’s an example of Microsoft creating role-specific Copilots (others being Copilot for Sales, Copilot for Service) that extend the core Copilot capabilities into specialised business functions.
Further Reading and Sources
As well my own experimentation, the following sources were also inferred and read when writing this blog. I did also use Copilot to help tweak the tone and flow.
Cisco Live 2025 is happening this week in San Diego (after five years in Vegas) with around 22,000 attendees. As you’d image from any tech event at the moment, the focus was very much AI with the theme being summed up as “All AI, all the time”. Throghout the Day 1 keynotes, Cisco’s message was clear: the “agentic AI era” is upon us, and Cisco is positioning itself as the infrastructure backbone to support service providers, cloud providers and enterprises of this new age.
Cisco’s President and Chief Product Officer Jeetu Patel set the tone with a bold analogy: “The way that you should think about us is like the picks and shovels company during the gold rush. We are the infrastructure company that powers AI during the agentic movement,”
…….In other words, while everyone’s chasing AI gold, Cisco’s approach is to providing the bedrock tools to dig for it – unveiling new innovations spanning networking hardware, unified management software, security, and collaboration tools, all infused with AI.
I wasn’t able to attend the event myself, but here’s my break down the top announcements and innovations from the live streams I watched. Let me know what I have missd 🙂
The “Agentic AI” Era
Cisco Live’s buzzword was undoubtedly “Agentic AI.” Cisco sees a shift from basic chatbots to autonomous agents that don’t just answer questions, but perform tasks and jobs on our behalf. As Jeetu Patel said in the keynote “The world is moving from chatbots intelligently answering our questions to agents conducting tasks and jobs fully autonomously. This is the agentic era of AI”.
Like many of the other tech giants, their view is that in this fast moving era, billions of AI agents could be working for us behind the scenes, which “will soar” the demand for high-bandwidth, low-latency and power-efficient networking in Cloud Providers and Private Hosted data centers.
Cisco’s key mesage here is that they are here to help organisations and providers meet this demand. “Cisco is delivering the critical infrastructure for the AI era — secure networks and experiences, optimized for AI that connect the world and power the global economy“.
Cisco CEO Chuck Robbins said that “no organisation can hire limitless people to tackle increasing IT complexity and cyber threats – instead – machines must scale to share the burden”. He went on to say how Automation and AI-driven operations are not just nice-to-haves; they’re becoming essential and every business is looking to invest and build here and it will only accelerate in pace and scale.
Cisco also set out to explain that “generative AI” and “agentic AI” have different effects on the infrastructure needed to support them. Generaive AI creates sporadic spikes in demand, but Agentic AI creates sustained perpetual demand for inferencing capacity. This means that for agentic AI, networks and Cloud data centers need a continuous heavy-duty upgrade to what they run on today. Cisco expect that many will large enterprises, those setting out to build their own “AIs” and of course Service and Cloud Providers will likley need to “re-rack the entire datacenter and rebuild the network” to handle these new AI workloads.
One Unified Plartform to Manage it all
As (a long time ago) IT Sys Admin, I remember how managing networks used to sometimes feel like herding cats – multiple dashboards for switches, routers, security, cloud, etc., all siloed.
Cisco has now announced Cisco Cloud Control, a new unified management console intended to “drive all its networking, security, and observability tools” from one place. In a nutshell, Cloud Control is Cisco’s approach to bring all those separate management tools into a single pane of glass – making it easier for network admins and giving a Cisco Customers a cohesive platform to showcase it’s new AI innovations in one place.
Of course Cloud Control is AI infused too. There is an AI Assistant that lets IT teams query their infrastructure in plain English. Here they could ask (as per their demo) “Hey Cisco, why is the Wi-Fi slow on the 4th floor?” and get a useful answer.
To achieve this, Cisco are using a new custom large language model trained on decades of Cisco networking knowledge (like an AI powered CCIE) to provide expert guidance. Cisco showed off a new AI Canvas (an “agentic” interface) that auto-generates relevant dashboards that work together to help identify issues, suggest fixes, and even implement changes – with human approval gating the final step. In short – you describe a problem, and the system brings forward the relevant controls and data needed to solve it, all guided by Cisco AI.
Cisco’s message is not just about adding AI for AI sake – it is designed to address real IT headache by combining formerly separate mnagement planes and interfaces into one.
Cisco also announced they are unifying management for their Catalyst and Meraki product lines (switching and wireless) into this single console, with common licensing too.
Overall, the message is that whether it’s campus networks, branch, data center, or cloud, Cisco goal is is to centralise control and inject AI assistance across them all, leading to smarter and simpler unified operations.
Splunk also got a mention – with Cisco talking about how ThousandEyes and Splunk analytics will also be able to integrate into this platform to give end-to-end visibility – from user device to application. This is part of a broader “One Cisco” vision of an integrated portfolio for networking, security, collaboration, and observability.
Net Hardware: Faster, Smarter, and Built for AI
It wouldn’t be Cisco Live without new hardware – and this year, Cisco delivered a loads of it. Recognising that AI workloads are putting unprecedented demands on Service provider and Cloud networks, Cisco unveiled a lineup of new switches, routers, and wireless devices which all give higher throughput, low latency, and security by design. This inlcuded:
Campus Switches (C9350 & C9610): Designed for campus networks and powered by its custom Silicon One chips – they boast a huge 51.2 Tbps of throughput and sub-5 microsecond latency, with quantum-resistant security built in. These are designed to handle “high-stakes AI applications” at the network edge.
Secure Branch Routers (8100, 8200, 8300, 8400, 8500 Series): To connect sites and users to AI resources, Cisco have unveiled these new Secure Catalyst Routers for branches. These are all-in-one boxes that combine SD-WAN, SASE (Secure Access Service Edge) connectivity and next-gen firewall. Cisco say they will deliver up to 3× the throughput of the previous generation too. Why? Cisco is converging networking and security at the WAN edge so that adopting AI doesn’t open new holes in your defenses.
Wi-Fi 7 (Cisco Wireless 9179F): – see new APs, tailored for stadiums and large venues. These APs support the latest Wi-Fi 7 standard bringing multi-gig speeds and better reliability and integrate Ultra-Reliable Wireless Backhaul (URWB) technology alongside Wi-Fi in one device. That means an access point can also serve as a highly reliable wireless bridge/mesh link, useful in places where running fiber/cable is hard.
Ruggedised Switches for Industry 4.0: To support AI at the edge – in places like factories, oil rigs, smart cities – Cisco unveiled 19 new rugged switches built to withstand harsh environments. These come in various form factors (tiny DIN-rail mounts, hardened casings, etc.) to fit into industrial sites where conditions are extreme. Interestingly, Cisco integrated that URWB wireless tech here too, meaning you can have a unified wireless fabric that covers both IT and OT (operational tech) environments via a combination of Wi-Fi and wireless backhaul. In plain terms, these rugged switches + wireless combos let factories and outdoor facilities achieve high-density, reliable wireless coverage as part of one unified infrastructure.
Powered by Cisco Silicon One: All Cisco’s hardware announcements reinforced a key point: networking and security are fusing together in Cisco’s strategy. All new switches and routers all come with baked-in security features (from Hypershield to post-quantum crypto) rather than treating security as an add-on. Jeetu Patel emphasised, that the future is about networks that are programmable and adaptable – Cisco’s own Silicon One custom chips are a big part of that story because it means that Cisco can update these devices for new AI workloads via software without needing to build a new chip and device. This is a major compete play and USP for Cisco.
Security in the AI Era: Zero Trust, Everywhere, All at Once
All the AI in the world won’t help if your business if your network isn’t secure. Cisco used this approach to double down on its message that security must be woven into every layer of the network, especially as AI opens new frontiers (and potentially new threats). In the agentic AI era, Cisco said that attackers will leverage AI, meaning threats could become faster and more sophisticated. The answer? “Secure by design” infrastructure and a unified security architecture that can handle the scale of AI-fueled operations.
As a result Cisco introduced a new network security blueprint anchored by what they call the Hybrid Mesh Firewall and Universal ZTNA (Zero Trust Network Access). They represent a concerted effort to integrate security across all users, devices, and applications more seamlessly including:
Hybrid Mesh Firewall: Annouced earlier this year, Cisco’s next-gen firewall for the AI era, acts as a distributed security fabric spanning your whole environment. It brings together Cisco’s own firewalls and even third-party firewall integrations into one cohesive system to to enable zero-trust segmentation everywhere – from your data center core, across clouds, out to branch offices and all the way to IoT devices at the edge. The goal is that every part of the network becomes a security enforcement point, tightly coordinated.
Universal ZTNA: Cisco’s Zero Trust Network Access solution, now branded “Universal” because it aims to cover any user or device, anywhere. Universal ZTNA provides secure, identity-based access to applications, whether users are on the corporate LAN, at home, or on a mobile device. It extends the zero-trust mode to hard-to-manage endpoints and ensures a unified policy follows the user. For example, whether JimBob from accounting logs in from the office or from a coffee shop Wi-Fi, the system continuously verifies his identity and device posture before granting access to the finance app. The synergy here is that integrating ZTNA and the distributed firewall, Cisco can tightly control user-to-app connections and even monitor the traffic between services, all under a zero-trust philosophy.
Beyond hardware, the cloud-based Cisco Security Cloud got enhancements to help secure those emerging AI workflows. Their platform can now better secure interactions involving AI agents, using tools like Cisco AI Defense (which monitors AI model operations for tampering or misuse) as part of a “Secure AI Factory” concept co-developed with NVIDIA.
Their integration of Splunk also got a mention, where they demonstrated deeper Cisco + Splunk integrations for security analytics – such as sending security events and network telemetry into Splunk’s SIEM and using Splunk’s AI-driven insights to automate responses via Cisco’s tools.
Webex: Smarter Meetings, AI Helpers, and Cameras with a Brain
Cisco did also announce a series of Webex updates with more AI coming into Webex in ways that aim to make meetings less of a chore and customer service more efficient.
Jira Workflow Automation in Webex: For native Webex meetings, this can listen for action items discussed in a meeting and automatically create Jira tickets for them. For example, if during a team call someone says “I’ll update the budget doc next week,” the AI will note that and generate a task in Jira , Monday.com or Asana – fill in your project tool) assigned to that person. It will even capture the context by attaching relevant portions of the meeting transcript or recording. Cisco touted, the integration can also update Jira tickets in real-time if status changes are mentioned in meetings – so, if the team says “the server migration is completed,” the AI could move the Jira task to “done” and note the discussion. It’s like having a diligent virtual project manager in every meeting, so humans can focus on discussion rather than note-taking.
Webex AI Agent for Customer Self-Service: They announced enhancements to the Webex AI Agent – to make it easier to deploy and more powerful. Tgherenis a new set of prebuilt, industry-specific templates – out-of-the-box chatbot templates tailored for industries like healthcare, finance, retail, etc. Instead of a generic bot that has to be trained from scratch, Cisco provides a starting knowledge base (e.g., a healthcare template might know common questions about insurance, appointments, privacy rules, etc.). This can significantly speed up creating a virtual agent and leads to more relevant answers since it’s contextually aware of the industry. Cisco are also enabling these AI agents and features for on-premises deployments as well.
Conclusion
Cisco is all-in on AI, not by making its own AI apps, but by supercharging the underlying tech that makes AI possible.
Cisco seem fully aware of the challenges businesses face with emerging technologies. – whether it’s handling the flood of data and compute that AI workloads generate, securing a more complex threat landscape, and having a true end to end view on the user experence – Cisco is positioning itself as the enabler (and problem-solver) and has signaled it’s not sitting on the sidelines of the AI revolution.
The narrative of “One Cisco” came through strongly: networking, security, collaboration, cloud, and services all interlinking to form a complete platform for the AI era. Cisco is offering a very compelling toolkit for enterprises: blazing-fast hardware to move AI bits, smart software to manage it with minimal hassle, and built-in security every step of the way.
Cisco wants to be “the infrastructure company that powers AI” – the dependable partner under the hood while everyone chases AI magic. By unifying its platforms and injecting AI into network operations, Cisco is making a play to stay indispensable in this new era.
One of the most frustrating thing about Teams intelligent Recap and Copilot in meetings is in its ability to not understand company acroymns and internal “language” or terms.
Scheduled to rollout in July 2025, tenant administrators will be able to upload a Custom Dictionary through the Microsoft 365 Admin Portal’s Copilot Settings page.
This feature will finally enables organisations to improve transcription accuracy in Copilot and Teams meetings and calls by enabling Microsoft 365 to understand company-specific terminology. This will means that will be able to understand things such as
Industry jargon,
Internal product names and terms
Multilingual terms
This should help ensure conversations are transcribed and interpreted with greater precision.
Why this matters?
Organisations rely on Microsoft Copilot and Teams transcripts for insights, documentation, and knowledge retrieval. However, standard AI transcription can misinterpret niche terms or acronyms, leading to confusion and even sometimes humorous transcriptions.
This new Custom Dictionary feature addresses this by allowing businesses to define key terms their workforce frequently uses.
Real Benefits.
Legal & Compliance Accuracy: Law firms using specialised legal terminology (e.g., “prima facie,” “voir dire”) can ensure precise transcripts without ambiguity.
Enterprise Acronyms & Branding: Technology companies like Cisilion will be able to maintain more accurate documentation of internal project names (e.g., “Project Nebula”) and proprietary solutions.
Global Team Collaboration: Multinational organisations can optimise transcription quality across multiple languages and regional dialects.
Better AI Insights & Search:Copilot will be able to retrieve knowledge more effectively, ensuring summaries, recommendations, and contextual responses align with an organisation’s unique vocabulary.
This update is part of a broader set of Microsoft 365 enhancements including improved accessibility for sign language users in Teams meetings and expanded Copilot capabilities for 1:1 and group calls.
By refining AI-driven language models, Microsoft aims to make workplace collaboration smarter, clearer, and more inclusive.
You can read more and track this features release on the official Microsoft 365 Roadmap.
There’s instructions for enabling and configuring it here.
Microsoft has just released its annual Work Trend Index report, and as anticipated, its focus is boldly centered on the transformative impact of generative AI in the workplace.
The report reveals that we’re on the brink of a paradigm shift where AI will not only reason but will also solve problems in unprecedented ways. Much like the industrial revolution or the dawn of the Internet, Microsoft suggests that a complete overhaul of work practices may take decades to fully materialise.
The annual Work Trend Index conducts global, industry-spanning surveys as well as observational studies to offer unique insights on the trends reshaping work for every employee and leader across more than data from 31,000 workers across 31 countries, LinkedIn labor market trends, and trillions of Microsoft 365 productivity signals,as well as leading AI-native startups, academics, economists, scientists, and thought leaders.
The full report is in the link below, but I’ve summarised the key insights around how the report claims generative AI is reshaping work and leadership dynamics.
AI’s Transformative Role
AI continues to advance in its reasoning and problem-solving ability, with the potential to revolutionise work.
Major transitions like the industrial revolution and the Internet took decades and view is that full wide scale AI may follow a similar path.
Immediate Impact of AI on work
Key Finding
Stats
AI Adoption
82% of industry leaders acknowledge AI is changing work
New Work Models
The “Frontier Firm” concept describes organisations using AI-powered intelligence on demand.
New emerging roles
The rise of “agent bosses”—professionals managing AI agents to enhance productivity.
Productivity Challenges
Challenges from the last 5 years continue to plaugue employees and impact productivity with tech overload and AI is seen as a “potential” to reduce this and bring better focus to information workers.
The report reveals that 80% of the global workforce feels overburdened by constant interruptions—an email, meeting, or ping every two minutes. Consequently, about 82% of leaders plan to harness AI and digital labor within the next 12 to 18 months to alleviate these pressures.
82% of leaders plan to harness AI and digital labor within the next 12 to 18 months to alleviate work and resource pressures.
Bridging Business Needs and Human
AI is making intelligence more accessible, shifting focus from headcount to on-demand expertise.
It helps close gaps between business demands and human workload.
Organisations are urged to invest in adoption training and business process reviews to determine the most optimum areas to leverage AI.
Key Finding
% impact
Executives concerned about productivity
53%
Global workforce feeling overburdened
80%
Average interruption rate
Every 2 minutes
Leaders planning to use AI to improve work
82% (within 12–18 months)
Talent, Hiring and Employment Trends
Addressing common concerns about AI replacing jobs, the report delves into LinkedIn data that indicates that top AI labs are hiring at twice the rate of traditional big tech companies. Interestingly, it says that much of this new talent is transitioning from established tech firms, underscoring a dynamic reshuffling of skills and expertise in the workforce.
The report also underscores the necessity for executives to strike the perfect balance between human talent and AI agents. As these digital assistants become ever more integrated into daily tasks, the role of the “agent boss” is emerging leaders who build, delegate, and manage AI agents to magnify their impact and steer their careers in the age of AI.
The report talks of a future where every worker, from the boardroom to the frontline, must adopt a CEO-like mindset for an agent-powered startup, predicting that within five years, 41% of teams will be actively training and 36% managing AI agents.
AI labs are hiring at 2x the rate of big tech firms.
Many AI hires come directly from established tech companies.
Human vs. AI Balance in Workplaces
Leadership Perspective
Employee Persective
67% of leaders understand AI agents.
40% of employees understand AI agents.
79% of leaders believe AI will accelerate careers.
67% of employees believe the same.
AI’s current and future role in Work Automation
AI Usage
Areas Impacted
46% of leaders use AI agents to fully automate workstreams.
Customer service, marketing, product development.
Organisations evaluating human-to-AI balance
AI integration must match societal expectations and business needs. |
But…..it states that AI is shifting the global work landscape, demanding strategic adaptation with..
Percentage
Areas Impacted
83% of leaders
believe AI will enable employees to tackle more complex tasks.
78% of leaders
want to recruit for new AI-related roles.
The report highlights that 67% of leaders are familiar with AI agents compared to only 40% of employees, and 79% believe that AI will accelerate their careers, outstripping the 67% noted for the broader workforce.
What other leaders are saying…
Bill Gates (Founder of Microsoft) said publically that AI might eventually perform “most things,”. We have also seen Salesforce CEO Marc Benioff already rethinking his company’s hiring strategies for 2025. This says that as we navigate this transformative wave, company leaders need to carefully consider when and what digital labour can complement, or in some cases, surpass that of human expertise, especially in roles that demand a personal touch or entail significant responsibility.
There is a necessity for executives to strike balance between human talent and AI agents. As these digital assistants become ever more integrated into daily tasks, the role of the “agent boss” is emerging—leaders who build, delegate, and manage AI agents to magnify their impact and steer their careers in the age of AI
Microsoft Work Trend Index Report
Summary
Microsoft’s Work Trend Index report paints a vivid picture of a future in which they show how AI is starting and has the potential to reshape every facet of our professional lives.
It claims that 83% of leaders believing that AI will empower employees to tackle more complex challenges and 78% actively looking to fill new AI roles, the global work landscape is poised for a dramatic evolution—one that necessitates a delicate balance between harnessing digital innovation and preserving the unique value that human insight brings to the table.
As March 2025 comes to an end, Microsoft have unveiled several exciting updates across Microsoft 365 Copilot, Copilot Chat, and Copilot Studio.
Copilotannouncementsthis week
1. Updates to Copilot Studio Message Rates
Effective April 2nd, 2025, updated (cheaper) message rates for Copilot Studio will go live. These adjustments cover tenant Microsoft Graph grounding and agent actions (previously known as autonomous actions). The prices of tenant Microsoft Graph grounding and autonomous actions are being reduced from 30 messages and 25 messages to 10 messages and 5 messages respectively, from April 2nd, 2025.
The following table illustrates the differences in the subscription models for the cost of Copilot Studio events.
Copilot Studio feature
Billing rate [non M365 Copilot Licensed users]
Billing rate [M365 Copilot licensed users]
Autonomous triggers1
Classic answer
1 message
No charge
N/A
Generative answer
2 messages
No charge
2 messages
Agent action
5 messages
No charge
5 messages
Tenant graph grounding for messages
10 messages
No charge
10 messages
Agent flow actions per 100 actions
13 messages
13 messages
13 messages
1 - Autonomous triggers refer to events or conditions that automatically initiate an agent to take action, without requiring a user to manually invoke it.
Also coming is Agent flows which allow agent creators to bring Power Automate automation features directly into Copilot Studio to quickly and consistently automate business processes. There will also be new deep reasoning in agents combines reasoning models including Open AI o1 with the ability to access enterprise data to complete complex tasks.
Microsoft are also updating pricing with a new zero-rating for Microsoft 365 Copilot licensed users in Microsoft 365 apps and services, ensuring inclusive, seamless integration and cost-effective use of these tools. This means licensed Microsoft 365 Copilot users will not be charged for using agents in their organisation
2. Rule-Based Workflows in Copilot Studio
From April 2025, Copilot Studio will introduce structured, rule-based workflows for agents. This aims to simplify process automation, enabling users to create efficient, consistent workflows with minimal manual effort. Usage of this functionality will contribute to the Copilot Studio meter, encouraging innovation while maintaining transparency in resource utilisation.
3. Deep Reasoning in Copilot Studio
So this is a big one – Microsoft have made deep reasoning capabilities available in Copilot Studio’s public preview from today. This will empowers users to address complex, decision-intensive tasks by leveraging advanced reasoning algorithms.
Whether it’s managing intricate processes or solving challenging problems, this tool offers remarkable precision and depth in its execution.
4. Two new Deep Reasoning Agents
Microsoft announced two new deep reasoning agents—Researcher and Analyst—as part of an early preview which will also be coming “soon” with previews coming in April before wider rollout.
Researcher Agent: has been designed for content creation and information synthesis, this agent combines OpenAI’s advanced deep research model with Microsoft Copilot’s orchestration. By integrating Copilot Chat’s web and work grounding capabilities, Researcher enables users to brainstorm ideas, generate high-quality content, and analyze data more effectively.
Researcher Agent in Copilot.
Analyst Agent: This is powered by a new reasoning model. the Analyst agent will function as a virtual data scientist and will have the ability to process complex datasets and provide real-time code validation (using Python) and will be able to deliver actionable insights and visually compelling representations of data in minutes.
Microsoft say that these agents will be gradually rolled out to Microsoft 365 Copilot licensed users through the Frontier program, an early access programme for customers to test out early and new Copilot innovations.
At the end of Feb 2025, Microsoft gave Apple Mac users with a brand-new native Copilot (consumer) app experience and now after a feeble Web app version, Windows 11 is finally getting a proper one too.
This latest update brings a fully native Copilot app to Windows, delivering a faster, smoother, and visually enriched interface that aligns perfectly with the Windows 11 design language. Yay.
It also has a keyboard shortcut that lets you hold the Alt + Spacebar keys for two seconds to start chatting to Copilot via voice.
From Web View to Native App
For those who followed the initial rollout, you’ll remember that the original Copilot for Windows was simply a web view of the Microsoft Copilot website. While functional, it left much to be desired in terms of responsiveness and overall polish.
Copilot App – Webapp to Native App
The new Copilot update transforms that experience completely. By leveraging the native app UI framework, Microsoft has infused the app with features that make the experience feel inherently Windows 11 that is also complete with a sidebar for managing chats, elegant mica blur effects, and native context menus and buttons.
This adherence to the native design not only improves aesthetics but also boosts performance and responsiveness.
What’s New in the Copilot for Windows App?
Enhanced User Interface
Native Design Language: The interface now mirrors the sleek, modern aesthetics of Windows 11.
Smooth Interactions: Launching the app is noticeably quicker, and interactions feel seamless thanks to the native integration.
Intelligent Chat Management
Sidebar for Conversations: All your previous chats are saved and easily accessible in a dedicated sidebar.
Instant New Chat: Starting a new conversation is as simple as hitting the new chat button.
Retained and Expanded Functionality
Text and Voice Chat: Continue to interact with Microsoft’s AI assistant using text, or opt for the Copilot Voice for a more dynamic experience.
Customisable Settings: Options include settings to enable or disable launching the app on Windows boot, as well as toggling the alt+spacebar shortcut for quick access.
In short, there’s no real feature changes here – just a native Windows App, ensuring that the native experience makes no compromises on capability and features along with performance and usability improvements of a native app.
First thoughts on the new version
I have to confess—I wasn’t thrilled with the old web view version of Copilot for Windows. It felt like an afterthought compared to its Mac counterpart. This new native experience, however, is a major improvement. The app now inspires confidence in handling everyday AI tasks and is genuinely enjoyable to use.
Getting the new Copilot App
For Windows Insiders excited to explore this update, the latest version (1.25023.107.0) or higher is now available via the Microsoft Store and should update automatically. The app is rolling out in preview across all Insider channels, inviting users to experience this transformative upgrade first-hand.
As a Microsoft product inside another Microsoft product, the evolution from a mere web view app (which should never have been done in my opinion) to a fully fledged native app that looks and feels like a Windows app not only elevates user interaction but also shows that Microsoft is actually serious about integrating AI seamlessly into everyday computing tasks.
The new Copilot for Windows app also has a keyboard shortcut that lets you hold the Alt + Spacebar keys for two seconds to start chatting to Copilot via your voice.
Microsoft want your feedback
Microsoft would like feedback too, which you can do by filing feedback in the Feedback Hub (WIN + F) under Apps > Copilot or directly within the Copilot app by clicking on your profile icon and choosing “Give feedback”.
This feedback shapes the future. Whether we can expect more iterative updates, possibly with additional features and enhancements will only happen based on the Microsoft collects feedback from Insiders.
Conclusion
The leap to a native interface is more than just a cosmetic upgrade—it represents a thoughtful stride toward a more integrated and responsive Windows experience. I’m excited to see how this native Copilot app will further inspire productivity and innovation as it evolves.
What are your thoughts on this updated native app?
Interesting article I read today about how the UK government is looking to change around the use of creative content without authors permission unless they choose to opt out.
The Government’s proposal is to change the laws to favour tech platforms, allowing them to use content without permission or payment unless the authors / creators specifically opt out.
Concerns
Creator and authors are challenging this arguing that this shifts the burden onto them. They believe that tech companies pay for using their content and training models based on their content.
They are urging the government to enforce copyright laws to ensure fair compensation for creators, securing the future of creativity and AI but in a way that protects the authors and creators of content.
The article calls for UK people to back this plan urging the UK government to enforce copyright laws to ensure fair compensation for creators and authors.
Where do you stand?
This is a tricky one for me as today, I feel most content LLMs are trained on are US based data sources and I would love to see more content based on UK data (after all I am British). At the same point, if a levvy is introduced and royalties paid a model for recouping costs is needed which may lead to this content being excluded in LLM training. This could lead to more bias and still leave them “out of pocket”.
In an new update announced on the Microsoft 365 Insider blog this week, Microsoft has announced that Copilot in Excel will soon be able to reference documents in Word, Excel, PowerPoint, and PDF formats jyst like the other officee apps can. This enhancement significantly expands the capabilities of Copilot in Excel, making it a more powerful tool for users.
With this update, you can now ask Copilot to perform tasks such as displaying to-do items in a table or organising emails with columns for the sender and subject line. This feature is particularly useful when you need to combine data from various sources, including public statistics from the web, internal documents, organisational details, or tables from another Excel files or contained in Word docs.
Getting Started with the New Feature
To take advantage of this new functionality, you need to meet the following minimum requirements:
Windows: Build 17729.20000 or later
Mac: Build 24053110 or later
Copilot license
Web search enabled
Stable internet connection
Upcoming Web Version and Limitations
Microsoft has announced that this update will soon be available for the web version of Excel. However, there are some limitations to be aware of. For example, refreshable data imports only work for Excel files with tables stored on SharePoint or OneDrive. Additionally, there is limited support for handling workbook and external data simultaneously.
Recent Updates to Copilot in Excel
Copilot in Excel has received several updates in recent months, further enhancing its functionality. One of my favourite features is the Clean Data feature, which addresses issues such as text and number inconsistencies.
Copilot has been integrated into the Excel start up experience, enabling users to use Copilot to explain what they want to create and receive improvement suggestions.
Looking Ahead: More Features on the Horizon
With Microsoft’s global AI tour taking place in cities around the world, including a stop in the UK on March 5th, we can expect even more exciting features to be announced soon. These updates highlight Microsoft’s commitment to continually improving Copilot and making it an indispensable tool for Excel users.
Stay tuned to my blog for more updates in Copilot and bookmark the Microsoft 365 Road map page.
Finally, it is here – Microsoft 365 Copilot now lets you save your prompts within Copilot for easy re-use later. Yes – this means you no longer need to save your prompts in separate documents or constantly copying and paste them.
When Copilot has completed its response(s), scroll back to your prompt in the chat.
Hover your mouse over the prompt – you’ll see bookmark and link icons appear.
5. Click on the bookmark icon to save the prompt to your library – you can also give it a friendly name to make it easier to find and reuse later.
Accessing Your Saved Prompts
Finding your saved prompts is just as easy.
Click on “View Prompts” above the chat box.
In the prompt library popup window, select “Your Prompts.” where you will be presented with a list of all the prompts you’ve saved.
Click on any saved prompt, and it will automatically paste the text into the chat window, ready for you to use again.
Why this feature matters
The ability to save and easily access prompts directly within Copilot enhances productivity and streamlines your workflow. It’s a small change with a significant impact, making it easier than ever to manage your prompts efficiently.
No more hassle, no more copying and pasting—just seamless, effortless prompt management.
OpenAI’s most advanced AI model “o1” which is known for its problem solving and deeper thinking has been available behind a $20 per month ChatGPT premium subscription. ChatGPT premium gives limited acess for $20 a month and unlimited access for $200 a month.
Copilot let’s you use it for free.
Microsoft has a tight partnership with OpenAI and is also on a mission to put their AI (Copilot) across every Microsoft Service it offers with huge capability and features even on theor “free” tiers.
Copilot Consumer Pro users have had access to Think Deeper (which uses the o1 model) for the past 12 months, but Microsoft have now made this feature free to everyone including those using the free version of Copilot.
To access it, you need to simply head ovee to Copilot on the web, (or via the mobile app) and ensure you are signed in with a Microsoft account (MSA). You then get completed free access to the Think Deeper search (which uses the o1 model).
How to get Microsoft Copilot
To get Copilot, head to the web (you actually find Copilot in the Edge browser) and go to https://copilot.microsoft.com or head over to you phones app store and search for Copilot and install it.
You need to be signed in with your Microsoft account to use these features.
Using o1 features aka Think Deeper
Once in Copilot, use the AI chat as you would before (or like you did in ChatGPT) and you will see a “think deeper” button inside the text input box.
Using Copilot’s Think Deeper (ChatGPT model o1)
Selecting it activates the o1 reasoning model. As it processed your prompt, you also get a spinning symbol since searches and responses using o1 are more thorough that with GPT 4 and typpically take around 30 secs.
Using Copilot’s Think Deeper.
This is Microsoft’s way of letting you know that you’re in for around a 20-30 seconds wait. If you don’t need deep search so for normal use), toggle this back off to go back to the super fast GPT-4o version…
So what can o1 do then?
The deep thinker feature of Microsoft Copilot is much better for more complex tasks and research due to the o1 model ability for in depth reasoning.
As such it is simply better for solving complex issues like math, logic or science, for analysing or creating long or richer documents and reports or for code creation and debug. The best way to test this is to run two Copilot Windows side by side and test out the same prompt with and without Think Deeper enabled.
Content created with o1 is also more “accurate” with far less AI hallucinations (aka, making things up).
Why do many GPTs Hallucinate? In general, GPT models learn by mimicking patterns in their training data (huge amounts of data). The o1 model uses a different technique called reinforcement learning, whereby it's language model works things out (though it's training) by rewarding the right answers and penalising wrong ones. This takes longer through the iterative and testing process. Once done the model moves through queries in a step-by-step fashion much like human problem solving.
o1 limitations?
It is worth noting that o1 isn’t quite on the same level as ChatGPT in some areas. It is less effective with factual knowledge and is currently less able to search the Internet and cannot process files and images.
What about DeepSeek?
The big story this week has of course been DeepSeek, a controversial Chinese AI firm that has announced and launched their own GPT-4 and o1 rivals that have been supposedly built at a fraction of the cost of OpenAI, Google and other US models, shaking share prices, disrupting the market and rasing many questions.
What is more is more is that DeepSeek models are claimed to be more advanced and faster than GPT-4o and smarter that o1.
The advent of DeepSeek has sent shockwaves through the tech industry. Global stock markets have reeled, sparking a cascade of investigations and looming threats of bans.
Yet, the bot hasn’t been without its champions. Interestly, Microsoft – OpenAI’s top financial invester and partner – has already embraced the DeepSeek R1 reasoning model, and has integrating it into Azure AI Foundry and also GitHub.
These platforms, beloved by developers for fostering advanced AI projects, now stand as the new playground for DeepSeek’s innovative potential.
DeepSeek logo
Open AI Strikes Back
In the wake of its free mobile app’s viral triumph, OpenAI’s CEO Sam Altman swiftly revealed plans to accelerate the rollout of new releases to keep ahead of its new Chinese competitor.
OpenAI are not standing still either though. Et the end of December 2024, month, they began trialing twin AI models, o3 and o3 mini. Remarkably, the former has surpassed o1 in coding, mathematics, and scientific capabilities, marking a significant advancement in their AI prowess.
There is no doubt this is an area that doesn’t stand still. By the time I click publish this post will likely already be out of date!
DeepSeek has certainly ignited an even greater sense of urgency within the already dynamic AI sector which moves and evolves on an almost daily basis.
Microsoft announced last week (15th Jan) that Microsoft 365 Copilot Chat is coming to every Microsoft 365 Commercial Customer regardless of whether or not they have paid Microsoft 365 Copilot licenses and what’s more we now get access to use agents with company data grounding support. Along with it comes a new pay-as-you-go tier that allows employees to access everything from chatbots to agents without the need for a Microsoft 365 Copilot license.
While Microsoft is still confident that the full Microsoft 365 Copilot remains “our best in class personal AI assistant for work“, the new pay-as-you-go tier means organisations can start using the technology at a much lower entry point and look to address key business cases rather than going full in on Microsoft 365 Copilot. .
“Copilot Chat enables your entire workforce — from customer service representatives to marketing leads to frontline technicians — to start using Copilot and agents today”. Jared Spataro | Chief Marketing Officer | AI at Work | Microsoft.
What is Microsoft 365 Copilot Chat?
Microsoft 365 Copilot Chat is Microsoft’s AI-powered chat feature designed to empower every person in every organisation to leverage Generative AI to make their “work lives easier and more efficient”.
For the employee, Microsoft Copilot Chat is a “personal assistant” they can chat with to get get answers, understand things better and get things done faster. Copilot Chat is It’s part of the broader Microsoft 365 Copilot suite but focuses specifically on enhancing communication and collaboration through chat.
How is Copilot Chat Different from Microsoft 365 Copilot?
The main differences between Microsoft 365 Copilot Chat and Microsoft 365 Copilot is three-fold.
Chat within Microsoft 365 Copilot provides work-grounded chat which means that Copilot can reason over data within your Microsoft 365 organisation such as files, SharePoint sites, your OneDrive, people (within Entra ID), your meetings, chat and email etc. Microsoft 365 Copilot Chat cannot access this data unless you “paste” into a chat window.
Copilot within the Office 365 Apps such as Outlook, Teams, Excel, Word etc is only available with Microsoft 365 Copilot.
Microsoft 365 Copilot is a paid add-on, where as Microsoft 365 Copilot Chat is included for free within your core Microsoft 365 licensing.
Microsoft Copilot Chat – Beyond Web Grounded Chat!
I’m personally not a fan of the name Microsoft 365 Copilot Chat because I do think it confuses people. The point I want to bring out here and why this was worthy of a post, is that previously, Copilot Chat (as it was called) only had access to data on the web and did note have the ability to leverage any of the new AI features such as Agents.
This has now changed. As the table above shows, with Microsoft 365 Copilot Chat, organisation will be able to create agents that do have access to data stored or connected to your Microsoft 365 tenant and also (and this is big) the ability for organisations to build and use autonomous agents (agents that can operate independently of a user).
The use of these new AI capabilities are paid for using a PAYG model. This means non Microsoft 365 Copilot users will have access to AI agents (for example in SharePoint) even if they themselves do not have a Microsoft 365 Copilot license.
What does Microsoft 365 Copilot Chat Provide?
Key Features of Microsoft 365 Copilot
Copilot Chat
Free, secure AI chat powered by GPT-4 and GPT-4o.
Ability to use Copilot Agents for automating tasks directly in the chat.
Support for file uploads in chat for summarising documents, analysing data, and suggesting improvements.
Support for Copilot Pages
Collaborate in real-time with AI and team members.
Integrate content from Copilot, files, and the web.
Create AI-generated images for campaigns and social media.
Agents
Ability to create and use agents using natural language to automate repetitive tasks.
PAYG / metered pricing for agents with IT control over deployment and management rather than requiring all users to have a Microsoft 365 Copilot license.
Copilot Control System
Enterprise data protection (EDP) for privacy and security.
Enables IT to better govern access, usage, and lifecycle of Copilot and agents.
Allows for measurement and reporting capabilities just like other Copilot Services.
Use Case Examples
A couple of use case scenarios are;
A customer service rep can ask a customer relationship management (CRM) agent for account details before a customer meeting.
A service or field service agent can access step-by-step instructions or real-time product information from information stored in SharePoint or Dynamics 365.
A sales person can get help with positioning a product based on information on solution propositions or marketing collateral.
How much does it cost?
Understanding the charges is not super straight forward to map. For comparison though, a Microsoft 365 Copilot license costs around $30 per user per month, so use this as a basis for comparison.
In another blog post, Richard Riley, General Manager of Power Platform at Microsoft said that “usage of agents is measured in ‘messages’ and the total cost is based on the sum of messages used by your organization.“
So what does that mean? Well, Microsoft now offers two ways for organisations to access the pay-as-you-go version of Microsoft 365 Copilot Chat:
Track each “message” sent to the AI whereby each message costs $0.01, billed monthly.
Pre-buying a pack of messages. This works a bit like a mobile data plan. As an example, you can buy 25,000 messages for $200 a month
The actual cost vary based on the type of response you need with responses that need generative AI costing more than responses that don’t.
Web-based answer: Free / no-cost
Classic answer: 1 message
Generative answer: 2 messages
Answers pulling data from company’s own systems (e.g., SharePoint): 30 messages
This paid capability is of course optional and organisations can decide whether to turn it ‘on’ or ‘off’ in Copilot Studio.”
Riley introduced the concept of “autonomous actions,” describing them as “generatively orchestrated triggers, topics, data connectors, and workflows, visible in the activity map displayed in generative orchestration mode“.
These are also available as pay-as-you-go, with a cost of 25 messages each time they act.
Here’s some costed use examples…
An agent answering customer questions online could use 500 classic answers and 2,000 generative ones, costing $45 for those 4,500 messages.
Another agent answering HR questions internally using Microsoft Graph data might use 200 generative and 200 tenant Graph messages, costing 6,400 messages or $64 for the day.
This approach allows businesses to fine-tune their AI usage to meet their specific needs, addressing concerns about the high costs of deploying these tools across enterprises. It also helps cost modeling certain scenarios much easier and provides an alternative to just giving every person a $30 per month Copilot License.
Using Microsoft 365 Copilot Chat
Assuming IT have enabled this in your environment, you can try this by navigating to https://m365copilot.com or by downloading the Microsoft 365 Copilot App from your preferred app store.
The integration of artificial intelligence (AI) into enterprise environments has introduced new security concerns. As adoption of AI continues at “cautious” pace, organisations must ensure the safety of the hundreds of AI apps that employees use (or try to use) sanctioned or unsanctioned as well as any AI applications built or customised by the organisation. This affects both data governance, exposure, and leakage as well as compliance.
I have aimed to not only compare their key features, similarities, and differences, but also to look at how both offerings can indeed help organisations based on specific business scenarios and needs.
Cisco AI Defense
Overview
Due to be released in March 2025, Cisco’s new AI Defense works slightly differently to Microsoft’s offering and is focused on securing AI applications throughout their entire lifecycle. AI Defense integrates with Cisco’s extensive network infrastructure portfolio providing specialised AI security measures.
Business and technology leaders can't afford to sacrifice safety for speed when embracing AI. In a dynamic landscape where competition is fierce, speed decides the winners. Fused into the fabric of the network, Cisco AI Defense combines the unique ability to detect and protect against threats when developing and accessing AI applications without tradeoffs". Jeetu Patel | Exec VP | Cisco.
Whilst not released yet, it will I have based this product release information I have read.
Cisco AI Defense focused on two primary areas of protection.
Accessing AI Applications: Recognising that whilst third-party AI applications can significantly boost productivity but may pose risks such as data leakage or malicious downloads. Cisco AI Defense is designed to give IT and SecOps full visibility into app usage and can enforce policies to ensure safe, secure access.
Building and Running AI Applications: Cisco acknowledge that developers require the freedom to innovate without worrying about vulnerabilities or safety issues in their AI models. AI Defense discovers your AI footprint, validates models to identify vulnerabilities, and applies guardrails to enforce security measures in real-time across both public and private clouds
Key Features
End-to-End Protection: Protects both the development and use of AI applications, ensuring safety and security throughout the AI lifecycle.
Network-Level Visibility: Leverages Cisco’s unmatched network visibility and control to detect and protect against threats.
AI Model and Application Validation: Identifies potential safety and security risks with automated vulnerability assessments.
Real-Time Protection: Offers robust real-time protection against adversarial attacks, including prompt injections, denial of service, and data leakage.
AI Cloud Visibility: Automatically inventories AI models and connected data sources across distributed environments.
Microsoft Defender for Cloud and AI
Overview
Microsoft Defender for Cloud and AI is designed to offer comprehensive security for AI applications and cloud services. Being a Microsoft product, it integrates seamlessly with Microsoft 365 and their wider cloud ecosystem, providing robust threat protection and security posture management. It also supports multi-cloud environments making it suitable for enterprise organisations.
Microsoft Defender for Cloud and AI’s primary protection areas are based upon:
Threat Protection and Security Posture Management: Microsoft Defender for Cloud and AI provides real-time threat protection for AI workloads and visibility into AI components, identifying vulnerabilities and offering built-in recommendations to strengthen security.
Integration and Continuous Monitoring: It integrates with Defender XDR for centralised alerts and continuous monitoring, ensuring security measures are enforced across hybrid and multicloud environments.
Key Features
AI Threat Protection: Provides real-time threat detection for generative AI applications, including data leakage, data poisoning, jailbreak, and credential theft.Real-time identification and mitigation of threats to generative AI applications.
AI Security Posture Management: Continuous monitoring and management of the security posture of AI applications, with automated vulnerability discovery and remediation recommendations.
Cloud App Security: Protection for SaaS applications, offering visibility into cloud app usage and protection against threats.
Prompt Evidence: Includes suspicious segments from user prompts and model responses in security alerts.
Extended Detection and Response (XDR): Integration with Defender XDR to centralise AI /workload alerts and correlate incidents for efficient incident management.
Integration with Microsoft Ecosystem: Seamlessly integrates with Azure, Microsoft 365, and other Microsoft security solutions and workloads.
Comparative Analysis
In short, both Microsoft and Cisco are providing products which complement their wider security portfolios to help customers better protect their organisations in the rapidly evolving world and adoption of AI technologies.
Similarities
AI Security: Both solutions focus on helping organisations secure AI applications and provide end-to-end visibility into their AI workloads.
Real-Time Threat Detection: Each offers real-time threat detection and protection, ensuring prompt identification and mitigation of security threats.
Integration with respective Ecosystems: Both solutions integrate with their respective broader security ecosystems (Cisco for Cisco products, Microsoft for Microsoft products).
Differences
Whilst both focus on security across the customers domain with a focus on understanding and protecting against (and keeping control of) AI based applications, there are clear, there are some subtle and unique differences.
Scopes of Use
Cisco AI Defense Specialises more in securing AI applications throughout their lifecycle including home grown developed services, where as Microsoft Defender for Cloud and AI is more focused on providing comprehensive security for both AI applications and SaaS applications.
Platform Integration
Cisco AI Defense provides deep integration with Cisco’s network infrastructure and other Cisco security products. Microsoft Defender for Cloud and AI has seamless integration with the wider Microsoft’s ecosystem, including Azure, Microsoft 365, Dynamics, Power Apps as well as being part of the wider Microsoft security solutions.
Capabilities
Cisco AI Defense places a key emphasis on AI-specific security measures that include automated vulnerability assessments and real-time protection against adversarial attacks.
Whilst similar in approach, Microsoft Defender for Cloud and AI offers broader security features, including threat protection for both AI and cloud services, and integrates with Microsoft’s XDR for centralised incident management.
When to choose which?
When to choose Cisco AI Defense
Best For: Organisations with a significant focus on AI development and deployment, particularly those heavily invested in Cisco’s network infrastructure.
Primary Benefits: AI model validation, runtime protection, and extensive integration with Cisco’s network and security products.
When to Choose Microsoft Defender for Cloud and AI
Best For: Organisations utilising a mix of AI and SaaS applications, especially those heavily invested in the Microsoft ecosystem (Azure, Microsoft 365, etc.).
Primary Benefits: Comprehensive threat protection, tight integration with Microsoft 365, Azure, Dynamics 365 and existing Microsoft security solutions.
Case Scenario: Ficticous Enterprise Organisation
Customer Profile: “A large enterprise organisation with a complex infrastructure, several hundred applications (mainly SaaS) as well as in-house and hosted custom applications running in Public Cloud (Azure), mix of productivity tools (Microsoft 365), AI-powered assistants (Microsoft Copilot and Chat GPT), multi-campus network environment (Cisco Meraki), Cloud Voice (Microsoft Teams), Space Management Tools (Cisco Spaces) and network performance monitoring (Cisco ThousandEyes).
Organisation has and uses Microsoft 365 E5. They have a contact centre based on Cisco Webex and use Microsoft Teams Meeting Rooms with Cisco endpoints. User devices as mix of Lenovo and Surface. They also use Cisco Duo. They have a Cisco EA.
They are in the middle of a Microsoft 365 Copilot pilot with around 20% of their organisation but aware that some other departments may have other shadow AI tools. They are also looking at building their own apps that will use a magnitude of AI agents and connectors.”
Cisco AI Defense vs Microsoft Defender for Cloud and AI
Given the complex infrastructure and diverse applications of this large enterprise organisation, the differences, strengths and similarities of each really stand out. Appreciating this a “made up” organisation, you can see where and why each product has its strength and merits.
Microsoft Defender for Cloud and AI
Given the extensive use of Microsoft services and the presence of Microsoft 365 E5, Microsoft Defender for Cloud and AI is highly recommended. It offers comprehensive security coverage for both AI applications and SaaS applications, integrating seamlessly with the existing Microsoft ecosystem. The core services are also included within the Microsoft 365 E5 subscription.
Key Benefits:
Broad Threat Protection: Covers both AI applications and cloud services, ensuring robust security across the organization.
Integration with Microsoft Ecosystem: Seamless integration with Azure, Microsoft 365, and the organisations other Microsoft applications and security solutions.
Centralised Management: Facilitates centralised management and monitoring, improving operational efficiency.
Cisco AI Defense
Considering the organisation’s significant investment in Cisco networking solutions and the presence of Cisco Meraki, Cisco Spaces, and Cisco ThousandEyes, Cisco AI Defense is also recommended. It provides specialised AI security measures and integrates well with Cisco’s network infrastructure.
Key Benefits:
AI-Specific Security: Focuses on securing AI applications throughout their lifecycle, providing tailored protection.
Deep Integration with Cisco Infrastructure: Enhances overall network security by integrating with Cisco’s network and security products.
Real-Time Protection: Offers robust real-time protection against adversarial attacks, ensuring continuous integrity of AI operations.
Combined Approach
Given the organisation’s diverse IT infrastructure and the need for comprehensive security, a combined approach using both Microsoft Defender for Cloud and AI and Cisco AI Defense is advisable. This dual solution ensures that all aspects of the IT infrastructure are covered, from AI applications to cloud services and networking.
By leveraging both solutions, the organization can achieve a robust, integrated security framework that covers all their IT needs, ensuring comprehensive protection and efficient management.
Budget and Management Considerations
Budget: While using both solutions might seem costly, the investment is likely justified by the enhanced security and centralised management capabilities.
Management: Both solutions offer centralised management, making it easier to oversee and control security measures. The tools are managed across the respective product suites which are already in use within the organisation minimising additonal admin / sec ops over head.
Conclusion
Cisco AI Defense and Microsoft Defender for Cloud and AI are both robust solutions tailored to different security needs and infrastructures. Understanding their strengths and integration capabilities allows organisations to make informed decisions, achieving comprehensive and integrated security frameworks.
Cisco AI Defense is new and will be available in March 2025, so please do let me know if I’ve missed anything obvious…
