11 Copilot Tips to Supercharge Your Productivity

Summer Of Copilot

The #SummerOfCopilot series rolls on – now in week twelve on LinkedIn. I  summarize my daily posts in my weekly Copilot Navigator newsletter, and offer a deeper dive here on the Insights blog.

There’s just one more issue in the Summer of Copilot newsletter to go after today. (It’s going to be a big one.) It may be the last week of summer in the Pacifici Northwest, but it's still a magical time.

Here’s where we are in the series:

#SummerOfCopilot Week 12 Blog Posts

• Week 1:  Trends and Insights from Microsoft’s Latest AI Innovations 

• Week 2:  First Encounters with Microsoft’s Copilot Researcher & Analyst 

• Week 3: From Tacit Knowledge to Talking Agents: Navigating Copilot’s Next Evolution 

• Week 4: Copilot Goes Ubiquitous – Agents plus Copilot training resources

• Week 5: Co-Piloting the Future – Custom AI Agents with Enterprise Control

• Week 6: AI in Action: How New Copilot Features are Changing Content Management and Productivity

• Week 7: The Future of Copilot – The Magnificent Seven

• Week 8: Driving Productivity: Copilot from Driver’s Seat to Board Seat

• Week 9: Copilot for Every Leader: How AI is Empowering the C-Suite

• Week 10: GPT-5 Meets Microsoft 365 Copilot

• Week 11: Your Copilot Preseason Playbook - 13 Wins

Our next post:

• Week 13: Copilot State of The Union – Looking Backward, Looking Forward

Over the past few weeks, I’ve shared some practical tips and tricks to maximize the value of your Copilot investment.

• Researcher, Facilitator, and People Skills Agents

• Prompting Tips - CARE/RAFT and Chain of Thought Prompting

• Using Instructions to standardize response styles and formats

• Excel’s new Copilot() function

• Custom Dictionaries and Brand Kits

• Video Creation

  
  

Whether you’re just catching up or looking for a refresher, here’s what you need to know – and how to take advantage of it. And first, we’ll look at recent Copilot news.

Copilot Updates

The summer of 2025 brought a surge of innovation to Microsoft 365 Copilot – so much, in fact, that it’s hard to keep up. Microsoft integrated OpenAI’s latest GPT-5 model into essentially all Copilot experiences, unlocking better reasoning and longer context in responses. They even confirmed plans to add Anthropic’s Claude models into Copilot alongside GPT-5, meaning Copilot will soon be multi-model. Why is that big news? Early tests show Claude excels at certain tasks – for instance, generating more polished PowerPoint slides from a prompt – so having both GPT-5 and Claude gives Copilot a broader toolkit. In practice, Microsoft will use a prompt router to dynamically choose the best model for your query.  They’ll also need to tackle the challenge of integrating Anthropic’s models to keep information contained in the M365 trust boundary.

On top of that, Microsoft rolled out a slew of new Copilot features and agents across Teams, Outlook, Excel, and beyond. If you felt like AI was creeping into every part of your workday, you’re not imagining it – from meetings to emails to analytics, Copilot got upgrades.

Agents

Use the Researcher Agent for Fact-Driven Insights

Imagine having a business analyst on call to dig up information – that’s what the Researcher agent is. It’s a Copilot mode where the AI goes beyond just your prompt: it actually performs research through your internal data and the web, then gives you a summary or report.

What it is: A built-in research assistant that can gather and synthesize info on any topic or question you have. It’s particularly good at pulling in data from online sources combined with your organization’s files, and presenting it with citations or references when available.

Why it matters: In business, time spent researching is time away from decision-making. Researcher agent can do in seconds what might take you hours – whether it’s market trends, competitive intel, or background for a client proposal. And because it integrates with your work context, it might surface that report buried in your SharePoint that you didn’t even know existed, alongside external stats. In short, it leads to more informed decisions, faster.

How to use it: You can explicitly add the Researcher agent to your chat. Or with GPT-5, just ask Copilot an open-ended question. For example, “Hey Copilot, what are the latest announcements from our competition that may have an impact on our GTM?” Copilot will recognize this as a research task (thanks to the new GPT-5 prompt router) and automatically invoke the Researcher agent. You’ll get an answer that might read like a mini report – key points, maybe bullet lists or paragraphs, often with references like footnote numbers. You can then follow up: “Can you show sources?” or “Give me more detail on point #2.” 

Pro Tip: Once you have the research output, you can copy it into a regular Copilot chat and prompt further, like “Summarize these findings in one slide for an executive.” The two-step approach (research, then refine) works brilliantly.

Pro Tip 2: Use Instructions to specify standard format preferences for research reports – long form paragraphs, short bullet pints, etc. More on that below.

Let the Facilitator Agent Run Your Meetings

Meetings have a life of their own – discussions flow, decisions get made, tasks assigned – and it’s hard to capture everything. Enter Copilot Facilitator.

What it is: An AI meeting assistant that actively participates in your Microsoft Teams meetings to record notes, highlight decisions, track the agenda, and even nudge the conversation if needed. It’s like having a project manager + note-taker in the meeting, but powered by AI.

Why it matters: It keeps everyone literally on the same page during the meeting. Instead of someone frantically typing minutes (or worse, no one doing it), Facilitator produces a shared set of notes in real time, visible to all attendees. This means people can correct or clarify points on the spot. It also means no one misses what was decided or who has an action item – it’s right there on screen.

How to use it: You can enable it before a meeting un Meeting Options (in Teams, not Outlook). (or IT) might need to enable it via Meeting Options when scheduling a meeting. Once enabled, Copilot will announce itself at meeting start (“Copilot is now recording notes and action items for everyone”). As discussion happens, you’ll see bullets appearing: e.g., “📝 Marketing strategy discussed – focusing on social media.” or “✅ Decision: Launch event in London on Nov 5.” Everyone in the call can see this live feed. You can also interact: say “@Copilot summarize this discussion so far” or “@Copilot, mark that last point as an action for Finance.” 

The differences from the regular Copilot in meetings are important: Facilitator’s notes are shared to all by default, whereas if you ask Copilot a private question in the meeting (the usual Copilot chat), only you see that.

Facilitator is also proactive – it might say “Time check: 10 minutes left, 2 agenda items remaining” even if no one asked. At the end of the meeting, you typically get an auto-generated recap (the Teams “Meeting Recap” feature) plus the structured notes that Facilitator captured. It’s a powerful combo to ensure nothing falls through the cracks. Many of our clients have started using Facilitator for key strategy meetings and report significant improvements in post-meeting alignment – everyone literally walks away with the same understanding of what happened and who’s doing what.

Leverage the People Skills Agent for Talent Insights

“Who knows about X in our company?” is a question that often goes unanswered or bounces around via email threads. The People Skills agent aims to answer that in seconds.

What it is: A new capability in Microsoft 365 Copilot that infers employees’ skills from their work and profiles, and a corresponding Skills Agent that lets you query and utilize that skills data. Essentially, Microsoft 365 is building an internal LinkedIn-style skills database for your org, auto-generated by AI. It looks at things like documents you’ve written, meetings you’ve participated in, emails, etc. (all the Microsoft Graph data) and uses large language models to identify “Jane is skilled in Data Analysis and GDPR compliance” or “Bob has expertise in JavaScript and Agile Scrum.” All this is done privately within your tenant – none of the data leaves or mingles with other companies, and there are controls for employees to opt out or edit their inferred skills.

Why it matters: This can dramatically improve how you assemble teams and develop talent. Leaders get a dynamic view of the “skill landscape” of the organization. For example, you might discover you have more machine learning expertise in a department than you realized, or spot a gap in cloud security skills in your IT division. It informs hiring and training priorities. And for day-to-day work, it helps you find internal experts fast – instead of pinging three people to find the SharePoint guru, just ask Copilot. It also empowers employees: people can discover colleagues with similar skills or find a mentor who has the skills they want to develop. Culturally, it encourages a knowledge-sharing environment because skills are more visible.

There’s also a big HR tech implication: as Josh Bersin highlighted, if Microsoft provides this “skills inference” natively, it could reduce the need for separate talent platforms. It essentially commoditizes something that vendors have been selling as high-end AI for HR. For companies, that’s potentially huge cost savings and improved data integration (since it’s all within M365).

How to use it: First, your IT admins need to enable People Skills (there’s an admin setup involving a skill taxonomy – Microsoft provides a built-in one with 16,000 skills sourced from LinkedIn’s taxonomy, which you can customize[). Once it’s running and has had time to infer (the initial indexing might take some time), you as a user will see new abilities in Copilot. For example, in Copilot Chat you could ask: “Copilot, who are some experts in our company on SAP integrations?” or “List people who have skill in machine learning and are in the Finance department.” The answer won’t list anyone who has opted out or marked their profile private for that info.

Leaders can generate a Skills Landscape report in Viva Insights which gives analytics like top skills in the org, or specific skill trends over time.

As an individual, you can ask Copilot about your own skills too: “What skills has Copilot inferred for me?” It might say (this is my own output, fyi):

🧠 Inferred Skills and Expertise

These skills have been consistently associated with your work, writing, and leadership:

🚀 AI Strategy & Deployment

• Leading AI adoption through hands-on, design-led workshops

• Building agentic AI use cases and governance frameworks

• Synthesizing global AI playbooks into strategic models 🧰 Copilot Enablement & Product Expertise

• Deep experience with Microsoft Copilot, SharePoint, OneDrive, Viva, and M365 launches

• Authoring internal enablement content

• Etc.

If something’s off, you can update your profile manually (the system gives individuals control – you can edit or remove skills, or opt out entirely from being included).

Your skills can also be mapped to relevant learning paths in Viva Learning, enabling personalized development plans aligned to current roles and future growth opportunities.

I’d encourage employees to view it not as “Big Brother” but as “boosting your LinkedIn without the manual work,” because having your skills known can lead to opportunities (just like having a complete LinkedIn profile can).

Prompting Tips

Master Prompt Writing with CARE and RAFT

We’ve all experienced the “Hmm, that’s not what I wanted” moment with AI. Nine times out of ten, the issue is the prompt. Writing effective prompts is a skill, and the CARE and RAFT frameworks are two easy formulas to get it right.

• CARE (Context, Action, Results, Example): This is about setting up your question. Provide Context – the background or situation. State the Action you want the AI to do – answer a question, provide ideas, analyze something, etc. Specify the desired Results or format – do you want a list of bullet points? a table? certain info included? Finally, give an Example if you can – humans do this in questions all the time (“for example, you could mention X”). It helps the AI understand the style or type of answer. Using CARE is like telling the AI: “Here’s what I’m dealing with, here’s what I need, here’s how I want it, and here’s roughly what I expect.”

• RAFT (Role, Audience, Format, Task): This focuses on shaping the answer. Assign a Role to the AI – e.g. “You are an expert project manager….” Specify the Audience who will consume the answer – e.g. “This is for a team of senior executives with no technical background” (this affects tone and detail level). Define the Format you want – maybe a memo, an email, a step-by-step guide, bullet points, etc. And clarify the Task – what outcome do you need (similar to Action in CARE, but RAFT’s Task is more output-focused). RAFT basically sets the voice and structure of the answer.

Why it matters: If you’re a business leader, you might not have time to coax an AI over multiple tries. CARE and RAFT help you get high-quality output in one shot. They reduce misunderstandings. Importantly, they force you to clarify what you really need – which often is half the battle.

I’ve seen prompts that are just one vague sentence get a mid response, but a prompt using CARE/RAFT (doesn’t have to be long, just more structured) yields an on-point answer that hardly needs editing. These frameworks also scale – you can train your team on them as part of “prompt engineering 101.”

How to use them: Let’s illustrate with a scenario. Say you want Copilot to help with an upcoming product launch strategy. A weak prompt would be: “Any advice for launching a new product?” 

Better with CARE/RAFT:

Context: Our company is launching a new line of eco-friendly cleaning products next quarter. We have a strong e-commerce presence but minimal retail footprint. Action: Provide a strategic plan outline for the launch, focusing on marketing and partnerships. Results: Give the answer as a bullet-point list grouped into ‘Pre-launch,’ ‘Launch,’ and ‘Post-launch’. Include 1-2 key metrics or KPIs to track in each phase. Example: For Pre-launch, an example might be ‘Engage 50 influencers for product reviews.’

Role: Act as a CPG marketing consultant. Audience: our executive team (familiar with marketing basics but looking for high-level direction). Format: at most 10 bullet points with short explanations. Task: Outline the launch strategy for our new line of eco-friendly products with timeline and metrics.” Whew, that seems long, but notice it’s just a few sentences covering each element.

The output from Copilot will be incredibly tailored: it might come back with something like

🟢 Pre-launch

Goal: Build awareness, generate anticipation, and prepare infrastructure.

• Influencer Engagement: Partner with 50 eco-conscious influencers for product reviews and unboxing content.

• Landing Page & Email Capture: Create a dedicated product landing page with early access sign-up.

• Content Strategy: Publish blog posts, videos, and social media teasers highlighting sustainability benefits.

• Retail Partnership Outreach: Begin conversations with 3–5 boutique eco-retailers for limited shelf presence.

• Logistics Readiness: Finalize packaging, fulfillment, and returns processes for e-commerce.

KPIs:

• Email sign-ups for early access: 5,000+

• Influencer reach: >1M combined followers

Etc.

Hint: we’ve got a free CARE/RAFT cheat sheet with examples across many different industries. So, print out CARE and RAFT, stick them to your monitor, (or just save them to your desktop) and watch your Copilot results improve.

Try “Chain-of-Thought” for Complex Problems

Sometimes you don’t just want an answer from Copilot – you want to see the thinking behind the answer, especially for multifaceted problems. This is where chain-of-thought prompting comes in.

What it is: It’s basically asking the AI to “show its work” by reasoning step-by-step. You can explicitly say “think step-by-step” or you might structure your prompt as a series of thought steps. Copilot (and GPT-5 in particular) will then produce an output that walks through the reasoning before concluding.

Why it matters: For complex decisions (e.g., “Should we enter a new market? Evaluate from financial, operational, and brand perspectives.”), a chain-of-thought response is incredibly useful. It’s like getting a consultant’s analytic write-up instead of just a yes/no or a simple list. It ensures the AI considers multiple angles. It also makes it easier for you to double-check the logic and trust the recommendation. Plus – this is key for leaders – you can often repurpose the reasoning steps as a first draft of a presentation or memo. It’s already structured logically! We find chain-of-thought especially helpful for strategy, planning, troubleshooting, and any scenario where the journey is as important as the destination. It can surface insights that a straight answer might gloss over.

How to use it: The simplest way is to add to your prompt: “Explain your reasoning step by step.” Or “Provide your answer in a sequence of thought stages, considering X, Y, Z.” 

For example, you might ask: “Copilot, what’s the best way to improve our customer service? Please reason it out step-by-step, including root causes of current issues and potential solutions, before giving a recommendation.” 

Copilot responds with a numbered list like (abbreviated):

🧠 Step-by-Step Reasoning

1️⃣Identify Root Causes of Current Issues

Several internal and external sources highlight recurring challenges in customer service:

• <..>

2️⃣Explore Potential Solutions

🔧 Operational Improvements

🤖 Technology Enhancements

🧩 Root Cause Analysis Best Practices

✅ Recommendation

To elevate customer service, Synozur should adopt a dual-track strategy:

📈 Track 1: Operational Excellence

🤖 Track 2: Intelligent Automation

You can see each step, and if something seems off (maybe Copilot assumed a cause that you know isn’t true), you can correct it and ask again.

Another trick: ask Copilot to consider a problem from multiple people in the chain-of-thought. “Think through this as if you’re the CFO, then the COO, then the Head of Customer Success.” You’ll get perspectives from each, which is great for rounding out your viewpoint. We’ve used this method in AI strategy workshops – seeing the AI articulate the pros and cons in a stepwise manner often sparks ideas among the human team, almost like having an outside consultant facilitate the brainstorm. Bottom line, when the question is complex, don’t just ask for an answer – ask for a thought process. You’ll be rewarded with depth and nuance.

Set “Standing” Instructions for Copilot (Your Personal AI Policy)

Ever get frustrated repeating yourself to an AI? For example, every time you start a chat, you have to say “please answer in a friendly tone and short paragraphs” or it forgets your preferred style. Microsoft heard that feedback and introduced Custom Instructions for Copilot (akin to the feature in ChatGPT).

What it is: A persistent set of instructions or preferences that Copilot will remember and apply to all future conversations (until you change them). It’s like writing a brief guide for how Copilot should behave with you.

Why it matters: Consistency, efficiency, and personalization. If you have a preferred format or tone, you set it once and Copilot’s responses will usually follow it – less editing for you. It also helps align with your company or personal brand guidelines. For example, if your company style is very formal, you can set that, so you don’t get casual emojis in every draft. On the flip side, if you want a bit of flair, you can tell it to be creative. Custom instructions essentially let you “dial in” the AI’s personality to fit your needs. Over time, as Copilot interacts with you, it might even refine how it applies those instructions (especially once the memory feature is fully integrated, as Microsoft is testing).

How to use it: In the Microsoft 365 Copilot interface, go to Settings | Personalization to find Custom Instructions. This feature started rolling out around July 2025. When you open it, you’ll have a text box to enter guidance.?” Here you can write things like: “Use a conversational tone, like a knowledgeable colleague. Always provide 2-3 bullet points if listing things. Never use slang. For emails, start with ‘Hi team,’ as a greeting.”

You can also state preferences like “Favor UK English spelling” or “Always include a call to action at the end if giving recommendations.” It’s very free form, so use natural language. Once saved, these instructions apply in all Office apps where Copilot is present (Word, Outlook, etc.). You can still override them by being explicit in a single prompt (your “on the fly” directions trump the standing ones), but otherwise they’re active.

Example from my own use: I use these:

• Most answers should have a 2–3 sentence introduction, a bulleted list of key points, and a 1–2 sentence conclusion (for quick readability).

• When drafting a report, use multi-level headings and full paragraphs, with the occasional graphic or suggestion for clarity.

• For a blog post, keep it under 1000 words and format all hyperlinks as descriptive inline clickable links.

• For a social media post (e.g., LinkedIn), start with an eye-catching one-liner, then use brief paragraphs or bullet lists (with emoji prefixes for flair). End with relevant hashtags, and put a single link in the first comment; suggest an image idea if appropriate.

This eliminates having to remind Copilot every time I draft a report to use headings and paragraphs, for example.

You can also create custom  instruction for each Copilot Notebook – so you can vary the instructions by project or information domain.

One more thing: If multiple people use Copilot in your org, each can set their own custom instructions.). But for most, it’s a boon. It essentially gives you a more tailored AI experience that gets better the more you fine-tune it. So don’t skip this – it’s worth spending 5 minutes on your custom instructions to save many 5 minutes in the future.

Try the New Excel =COPILOT() Formula

Excel just got a lot smarter. If you’re an Excel user, you know formulas like SUM() or VLOOKUP(). Now there’s a formula that brings AI into the mix: COPILOT().

What it is: A function you can enter in a cell that takes a prompt (and optional cell/range references) and returns an AI-generated response right in your spreadsheet. It’s currently in testing (Beta channel) for M365 Copilot users as of late August 2025.

Why it matters: This bridges the gap between unstructured tasks and structured data. You can manipulate and analyze text or get insights without leaving Excel. It’s great for tasks like summarizing a column of feedback, categorizing free-form text, or even generating synthetic data for a scenario. And because it’s part of Excel, the output can be used in charts, combined with other formulas, etc., like any other cell value. It also updates live: if source data cells change, the function can re-run (within limits) to update the result. Think of the scenario: you maintain a project status spreadsheet – now you can have a cell that says =COPILOT("Summarize overall project health based on the statuses in column B") and always have a fresh summary each week. That’s powerful.How to use it: Assuming you have access (requires a Copilot license and Excel), you just enter it like a normal formula:

=COPILOT("Prompt here", [optional reference1], [optional reference2]...)

The prompt is in quotes like a text string. You can intersperse references. If you have a prompt that’s static plus a cell value, you can concatenate, or use multiple arguments. For example:

=COPILOT("Using the quarterly sales in ", B2:B5, ", write a short analysis.").

Or a simpler one:

=COPILOT("Classify each feedback as Positive, Neutral, or Negative", A2:A100).

That formula, entered as an array formula, could spill the classification for each of the 99 feedback entries into the cells below it (because it’s one prompt for many outputs).

Excel batches some calls to the AI to be efficient. Excel also currently limits calls to 100 every 10 minutes. If you drag a COPILOT formula down 100 rows, that’s 100 calls – hitting the cap. But if you instead refer to a range and output an array, that’s 1 call. Excel nerds will optimize their usage accordingly.

You can also nest COPILOT with regular formulas: e.g., =IF(condition, COPILOT("do X"), COPILOT("do Y")) – though that might be two calls if condition differs by row.

As this feature rolls out to general availability (later in 2025), expect to see a lot more “smart spreadsheets” in your organization. If you’re in finance, ops, or any Excel-heavy role, keep an eye on this – it could become your favorite new function.

Improve AI Accuracy with Custom Dictionaries

Have you ever read a Teams meeting transcript or Copilot summary and stumbled on a garbled proper noun? Like a project codenamed “Apollo” transcribed as “a polo,” or a client name like “Schneider” coming out as “Snyder”? I’m always surprised that Microsoft’s own “Azure” usually comes out “As You’re” It’s a minor annoyance that can create confusion or just look unprofessional. Now multiply that by all the unique terms in your business. Custom dictionary for Copilot is the answer.

What it is: A feature that lets organizations feed Copilot (and Teams transcription) a list of custom vocabulary to recognize. This feature rolled out around August 2025 in the Teams Copilot settings.

Why it matters: It reduces errors in transcripts and AI responses, which means less cleanup for you and higher confidence in AI-generated content. When Copilot knows “Synozur” is a company name, it won’t accidentally turn it into an English word or something. This is especially crucial for industry-specific jargon, acronyms, product names, or any proper nouns. Imagine a pharma company with drug names – you definitely want those right when Copilot summarizes meeting notes about them. Also, when transcripts are accurate, Copilot can better use them for context in answering questions (since it references the transcript text). It’s a quality-of-life improvement that greases the wheels of all other Copilot features.

How to use it: This one is admin-driven. A Teams administrator (or Copilot admin) can go into the Microsoft 365 Admin Center > Copilot settings and find Custom Dictionary. There, they can download a template and then upload a CSV with names, pronunciation, and full names. it at least ensures the speech-to-text engine in Teams recognizes those terms. For Copilot’s text generation, having those terms in its vocabulary means it will use them correctly in responses (and not flag them as unknown).

Example: We at Synozur added terms like “Synozur” (our company), “OKR” (the acronym, so it doesn’t become “Oakley are” or “Oh, Care!”)product names of our clients that come up often, and some internal project codenames. Immediately, our meeting recaps looked cleaner. The custom dictionary can also include acronyms – which is handy because Copilot sometimes tries to expand acronyms

incorrectly.

This is a quick win: a small setup effort for IT, but a noticeable improvement for everyone. If you’re the champion of Copilot at your org, consider nudging your admin to implement a custom dictionary. Compile a list of the top 20-50 unique terms in your corporate lexicon and get them uploaded. Your colleagues might not realize it, but their AI just got a bit more fluent in “speaking your company.”

Manage Your Copilot Agents (and Pin the Best Ones)

As organizations jump on the AI train, many will develop custom Copilot agents – think of them as chatbots or AI assistants tailored to specific needs (HR policies bot, IT help bot, Sales playbook bot, etc.). Microsoft’s Copilot platform allows these to be built (using Copilot Studio) and deployed in your tenant. That’s great – but soon you might have dozens of them floating around.

What it is: Microsoft introduced an agent management system in the M365 admin center that provides an inventory of all Copilot agents in your environment, plus governance controls like approval workflows and usage metrics. Additionally, they enhanced the user experience by making agents easier to discover (for example, SharePoint-based agents showing up in the Teams app store and Copilot app sidebar).

Why it matters: Without this, companies could face “bot sprawl” – a chaos of redundant or outdated agents, some possibly with inaccurate info, and users not knowing which one to use. Or "Hello World" demo bots that are still running and incurring consumption costs.

The new features ensure AI solutions remain high-quality and trusted. It puts IT in a role to curate and guide, which is important for adoption. If employees stumble on a poorly made bot that gives wrong answers, they’ll lose trust in the AI initiative. Conversely, if they are presented only the vetted, useful agents, they’ll use them more and get more value. Governance also covers risk: for instance, if someone built an agent that inadvertently exposes sensitive data, admins can catch and correct that. It’s akin to an app store for internal AI – with curation.

How to use it: For admins: head to the Copilot section in the admin portal. There, under Agents, you’ll see all agent instances: Microsoft-provided ones (like the global Researcher agent), third-party (if any integrated via connectors), and your internal ones (those built by employees or partners for your org). You can click each to see details: who published it, usage stats, and actions to take.

New in the summer updates is the ability to block or retire agents right from that interface – for example, if the employee who made an agent leaves, you might retire that “ownerless” agent to prevent an orphan bot situation.

There’s also an approval workflow: if someone builds an agent in Copilot Studio and wants to share it broadly, it can require an admin approval first. The admin gets a notification (in the “Requested agents” list) and can review what the agent does, test it, then approve or reject. If approved, that agent can be published for everyone or a specific group.

Now for users: Microsoft has made it easier to find these agents. One way is through the Microsoft 365 Copilot app (the central Copilot interface, kind of like a hub). It now has a sidebar or menu for Agents, including those you have access to and ones you recently used. It even automatically lists “recently used” and potentially allows pinning favorites.

Turn Your Docs and Slides into Videos (with Copilot Create)

We all know the saying: “a picture is worth a thousand words.” What about a video? Copilot helps you create videos out of existing content.

What it is: A capability in the Microsoft 365 Copilot Create mode (augmented by the Clipchamp video editor) where you can input a prompt, document, or PowerPoint and get an auto-generated short video with narration, subtitles, and visuals. Think of it as AI video summarization.

Why it matters: Video can engage an audience in ways text or slides might not. But producing even a 2-minute video can take skills and time that many teams don’t have. This feature mainstreams it – now anyone who can describe what they want (or provide a slide deck for reference) can get a draft video. This is particularly useful for internal communications, training, marketing teasers, or social media content.

How to use it: In the Copilot interface, go to Create ->Create A Video. You need to describe the video you want and/or provide a file to base it on. If describing, you’d say e.g., “Create a 1-minute video summary of our Q2 performance. Use a serious tone. Highlight the 3 biggest achievements and include our company logo.” Copilot will then generate a script and fetch relevant visuals. You’ll get a preview of scenes, each with suggested imagery (maybe from a stock library or from your own slides if you provided them) and the voiceover text. If you provided a PowerPoint, Copilot will parse the slides for key points and use any images on them to create the scenes

Once Copilot creates the draft video, , you can edit it -  the text of the narration, the brand kit, etc. – in Clipchamp.

Keep Everything on Brand with Brand Kits

Last but not least, let’s talk about brand consistency. Enterprises invest heavily in their brand identity – logos, colors, fonts, tone of voice – and understandably they don’t want AI content going off-script. Microsoft addressed this with Brand Kits in Copilot.

What it is: A feature in the Copilot Create experience where organizations can upload their official branding assets and guidelines, which Copilot will then apply to all relevant outputs. In practical terms: you have a Brand Kit with your company logo, brand colors, preferred fonts, perhaps even templates (like a PowerPoint master). When Copilot generates something visual (a poster, an image, a flyer, etc.), it uses those assets by default. Even for text, the Brand Kit can carry a defined tone/voice style that the AI will try to follow.

Why it matters: It ensures quality and consistency. Without it, if someone uses Copilot to make a presentation, they might get some random look – wrong colors, etc., which then someone has to fix. With Brand Kits, the first draft is much closer to final. This saves time and prevents “off-brand” materials from floating around. It also effectively extends your brand governance into AI-generated content. Companies have spent years training employees (and agencies) to follow brand guidelines; now AI tools need the same training. Brand Kits give you that control. For the employees, it means less tedious formatting – they can focus on content, knowing Copilot will make it pretty and compliant with the style.

How to use it: An admin sets this up. There’s an M365 policy setting to designate a mail enabled security groups  as brand managers who can publish official kits.

They then go into the Copilot Create interface, and create a New Brand Kit. They’ll upload logos (in various formats), specify brand colors (hex codes), choose fonts (Copilot will have those fonts available to use in outputs), and possibly set default styles for things like charts or illustrations. There’s also usually a place to write a short Brand Profile – a description of your brand’s personality/voice. For instance, “Contoso’s voice is friendly, inclusive, and professional. We use humor sparingly. We refer to ourselves as ‘we’ and the customer as ‘you’.

Once the kit is saved and published, users can select it when using Copilot.

In our use: I created a Synozur brand kit in Copilot. It includes our two logos (light and dark background versions), our color codes, and our chosen font (Avenir Next LT Pro). And it’s published to everyone to help output stay on brand with polish and consistency.

(Hint: You can also decide not to use a brand kit sometime – you may want an image in regular color, not your brand colors, for example.)

From Summer Experiments to Everyday AI

This summer showed us what’s possible when powerful AI meets daily work tools. Microsoft 365 Copilot evolved from a useful assistant to an entire ecosystem of AI helpers – we now have agents that can attend meetings, comb through data, understand our people, and even enforce our branding. The tips above are not just isolated tricks; together they signal a new way of working.

As a business leader, what should you do with this knowledge? Start integrating these capabilities into your team’s workflows, one step at a time. Maybe begin with meeting productivity – turn on Copilot’s Facilitator for your next staff meeting and feel the difference. Or if your strategy team is spending weeks on research, have them give the Researcher agent a try to gather intel. Encourage your people to practice writing better prompts (CARE and RAFT can even be a lunch-and-learn topic). Small improvements in how you interact with AI can yield big productivity gains. And don’t overlook the governance pieces: set up your Brand Kit, tune those custom instructions, and have IT keep an eye on the agents being created. That ensures the AI wave doesn’t become the Wild West.

It’s also a good time to update policies and training. For instance, with People Skills, you might want an internal guideline on how employees can use and update their skill profiles – to get the most benefit out of it. Or with the Excel COPILOT function, perhaps some best practices on verifying AI-generated analysis (we always stress: trust, but verify, especially for critical decisions – AI doesn’t remove the need for human judgment).

Stay tuned – Copilot is evolving quickly and new patterns and good practices emerge almost daily. Thanks.