Ep 621: Microsoft’s new Agent mode: Microsoft marketing or Future of Work in Windows?

Microsoft’s New “Agent Mode” and “Office Agent”: Timelines, Limitations, and Enterprise Value Unpacked

As headlines swirl around Microsoft’s latest AI releases—Agent Mode and Office Agent—many are left wondering whether these features signal the next step in workplace automation for Microsoft 365 users, or simply lend themselves to surface-level buzz. Directly referencing the product rollout, access specifics, and practical implications unpacked in the recent Everyday AI podcast, here’s a breakdown of what business leaders, IT decision-makers, and enterprise users need to know.

What Microsoft Is Actually Releasing: Understanding Agent Mode vs. Office Agent

Microsoft has launched two distinct tools aimed at streamlining productivity for users of Microsoft 365 Copilot:

  • Agent Mode: Embedded within Excel and Word, currently only for web versions, Agent Mode allows for iterative, multi-step task execution inside documents and spreadsheets. It leverages “OpenAI’s latest reasoning models” (exact version not disclosed) to assist with planning, executing, evaluating, and refining outcomes. Notably, it can handle autonomous workflows such as financial modeling or multi-file document updates, iterating until results are verified.

  • Office Agent: Available exclusively (for now) via Copilot Chat—not directly in Office apps—Office Agent is designed for “chat-first” creation of artifacts like Word documents and PowerPoints. It conducts clarifying Q&A, runs rigorous web research, and produces research-backed outputs, featuring live previews and continuous quality checks. It runs on Anthropic’s Claude 4.1 models (Sonnet and Opus specifically).


Who Gets Access—and Why That Matters

Despite impressive demos, access is highly fragmented and, at launch, counterintuitive for core business and enterprise users:

  • Agent Mode is available to holders of Copilot licenses within both enterprise and personal/family Microsoft 365 subscriptions, but strictly as a web-only feature for now. In Excel, users must add via the “Excel Labs” add-in, a step that diverges from Word’s more straightforward toggle integration.

  • Office Agent—the more publicized, versatile tool—is currently restricted to U.S.-based personal and family Copilot subscribers. Notably, it is unavailable to enterprise customers at launch, despite being tailor-made for automation-hungry organizational environments. Enterprise availability is planned but not yet scheduled.

This tiered and staggered rollout creates a paradox where large businesses possessing the bulk of Microsoft’s Office licenses are unable to utilize the highest-profile features until further notice.


Workflow Impact and Concrete Capabilities

Several capabilities stand out for organizations once these features are generally available:

  • Agent Mode in Excel: Enables high-volume data manipulation, such as handling financial models, creating dynamic loan calculators, and running complex data visualizations across large datasets—directly via natural language prompts. Previously, data practitioners were constrained by single-turn assistive tools and cumbersome manual formula searches.

  • Agent Mode in Word: Focuses on orchestrating structured drafting, automated updates across multiple files, and consistently refined formatting—again, through iterative dialogue rather than shallow one-off suggestions.

  • Office Agent in Copilot Chat: Delivers multi-step research-backed assembly of business artifacts, including in-depth PowerPoint decks tailored by presentation length, style, and structure. The agent’s guided clarification phase significantly reduces back-and-forth and minimizes ambiguity in generated outputs. Integration with Claude allows for thorough web research, followed by content generation, all with stepwise transparency and hands-on modifiability (“remove image in slide six,” “change bullets to visuals”).

The biggest differentiation from prior versions is this multi-step orchestration—breaking down end goals into actionable subtasks, autonomously iterating based on user feedback, and validating output without rebooting the process from scratch.


Bugs, Bottlenecks, and User Experience Puzzles

Several concerns were raised regarding usability and rollout:

  • Performance: For advanced, resource-intensive tasks (such as Excel data handling), Agent Mode often outperformed other public LLMs in completeness and accuracy, but noticeably lagged in speed. Iterative correction (such as fixing cell merge bugs) can be slow, delaying adoption for time-sensitive professionals.

  • Confusing Access Paths: The way Agent Mode is engaged differs between Excel (Excel Labs add-in, web-only) and Word (direct toggle), while Office Agent sits behind a Copilot Chat interface. This inconsistency complicates both IT administration and adoption, especially for organizations managing large user bases.

  • Fragmented Permissions: Multiple anecdotes underscore organizational confusion about who should use what, where, and how—meaning that, while Copilot agents are technically accessible in many places, actual business usage often lags due to permission mismatches and ambiguity over the correct workflow.


Marketing vs. Meaningful Adoption: The “Vibe Working” Problem

Microsoft’s promotional campaign leans heavily on the “Vibe Working” concept—a reframing of workplace creativity and productivity—but this approach misses the mark in the context of current product access and user experience. With most business users still struggling to navigate Copilot’s inconsistently rolled-out features, the risk is that these new agents are perceived as vaporware or superficial upgrades.

Enterprise leaders should be aware that, when features like Office Agent and Agent Mode do reach broad availability, they will likely deliver transformative value in document and data-centric workflows. However, organizations should prepare for a steep adoption curve, built around clarity in deployment, robust training, and streamlined access protocols.

Immediate Actions and Strategic Preparation

Given these constraints and opportunities, organizations should:

  • Monitor Rollout Schedules: Track Microsoft’s phased release plans for enterprise enablement, and communicate updates internally to relevant teams.

  • Experiment with Available Features: Where possible, test Agent Mode in web Excel/Word, via personal or family Copilot subscriptions, to document user experience and build local expertise ahead of wider rollout.

  • Prepare for Integration Complexity: IT and operations teams will need to adapt onboarding and usage policies to accommodate inconsistent feature placement and evolving access requirements.

  • Benchmark Outputs: Cross-validate agent-driven artifacts (especially in Excel) with manual workflows to ensure accuracy before mission-critical adoption.


Conclusion

Microsoft’s new AI agents represent a substantive leap in document and data automation—when and where accessible. However, the current rollout is ammunition for business leaders to scrutinize deployment schedules, clarify user permissions, and prepare a structured adoption path. The fragmented release may frustrate some today, but eventual enterprise integration will likely offer significant productivity gains in highly repeatable, document-based business use cases.

For those responsible for technology strategy and modernization, now is the time to chart adoption, pilot new features, and ready teams for a smarter, if presently uneven, future built on Microsoft’s maturing agentic tools.


Podcast Transcript


Jordan Wilson [00:00:16]:
Microsoft just released two new agents, agent mode and their office agents. And on the surface, man, in the demo videos, they look pretty good and extremely useful. So are these new agents, the future of work, if you're a windows user and Microsoft three sixty five Copilot user? Or is this just another case of Microsoft's marketing looking really good and millions of people won't know how to use these agents? We're gonna dive in and hopefully find that out today on everyday AI. What's going on y'all? My name is Jordan Wilson and welcome to everyday AI. This is your daily live stream podcast and free daily newsletter helping everyday business leaders like you and me, not just keep up with what's happening in the world of AI, but how we can make sense of it, grow our companies and our careers. If that's what you're trying to do, it starts here with the unedited, unscripted livestream podcast. But if you wanna take it to the next level, make sure to go to our website, youreverydayai.com. There, we recap each day's podcast in our free daily newsletter, and there's also, like, 620 plus episodes you can go listen to by category, watch the video, all that good stuff on our website.

Jordan Wilson [00:01:35]:
So, yeah. And AI news went straight up bonkers today. Y'all, I mean, we we heard about, you know, Sora too. We got Claude, four or five, you know, these new agents, right? So much happening in AI news today, so make sure you go check out the newsletter. But we're here to talk about Microsoft's new agent, their agent mode, and office agents. Let's be honest. They look great. They look great.

Jordan Wilson [00:02:03]:
But the real question is, will anyone be able to actually use them and access them? And it's hot take Tuesday. So I've got some takes on that. But the first thing is this is you guys wanted this, right? So, you know, if you are a long time podcast listener and maybe some of the episodes you're like, hey, dude, this stinks. Well, go sign up for our newsletter. I do a lot of polls in the daily newsletter asking you all what you wanna hear. So, if you want to hear more of something, well, go make sure you read our newsletter and vote in the polls, but overwhelmingly you all even more so than the new chat GPT pulse, which I thought everyone would want to hear about. But you all want to hear about the new copilot agent mode. So, today we're gonna go over Microsoft's announcement of agent mode and their office agent, tell you who can use them, how and what work they can start to automate.

Jordan Wilson [00:02:53]:
And I'm gonna tell you if this is actually a fundamental shift in how millions of Windows workers are gonna work, or if it's just some cool marketing and a nice little feature that may or may not be useful. Let's get straight to it. I won't make you wait any longer. My hot take, Well, Microsoft made a huge marketing mistake here. They're leaning into this thing called Vibe Working, which seemed kind of out of left field. And when we dive into these, two new agents a little bit more, you'll probably see what I'm talking about. Is this true vibe working? Not really. Is Microsoft in prime position to maybe be the vibe working platform of the future? Probably, but I don't think this is it.

Jordan Wilson [00:03:39]:
Also, this is extremely useful for enterprises that have all of their data in line. That's who I think this is ultimately for, but also not really. Anyways, if if you are an enterprise company with your data in Microsoft March, this is will actually change how you work. Well, once everyone gets access, because that's another thing. I think Microsoft, sacrificed clarity and user experience in just chasing some new agents that they can put on their website and say, we release these new agents. Well, there's a lot of, things that I think just were bad. Right? So, at least right now, they're web only. There's no commercial accounts for some of them.

Jordan Wilson [00:04:24]:
So let's just even just take a look at this. So the agent mode, there's a different agent mode in Excel and a different agent mode in Word. But right now, those are online, but they are for both, kind of personal. So personal and family subscribers. So you have to be a Microsoft three sixty five Copilot subscriber. So you have to have obviously a Microsoft three sixty five license, a Microsoft three sixty five Copilot license. But to use agent mode in word or Excel, you can have either a normal commercial type of license, or a family or personal subscription. Here's what's weird.

Jordan Wilson [00:05:04]:
Office agent. What the heck? This is the one I was most excited about. In at least right now, it's only available to personal and family subscribers. It's not available to licensed customers, which makes absolutely no sense. Yes. This has gotta be the hottest thing of all like Microsoft. Why? Right? When you release something called office agent, you know, you're known for office and the majority of your enterprise customers can't use said office agent. Come on.

Jordan Wilson [00:05:42]:
Like you don't need 220,000, employees to know that that's a terrible mistake. They shouldn't have done this. This is going to make, what is already a convoluted copilot experience even more, full of webs to untangle. I spent honestly three hours trying to understand how different groups of people get access to the different agents and how it is extremely confusing, which is unfortunately on brand for Microsoft Copilot, which is sad because once their products, once you get them to work, and you can give everyone access, Copilot's great. Right? I think unfortunately, especially in the AI circles where us AI dorks are running, Copilot doesn't get a lot of credit, although I think they should, you know, especially now since they're starting to use the GPT five models from open AI, they just started rolling out, Claude models. If we're going to be talking about that here in these agents as well. So Microsoft should be getting more credit, but the user experience is terrible. Getting people access, to at least these agents is is gonna be disjointed, and it's gonna be such a a delayed rollout, it seems.

Jordan Wilson [00:06:57]:
So you'll only get one big shot if you have come out with something called agent mode or office agent. And I think Microsoft squandered it. Sorry. All right. But let's talk about, these two new agents. So this new there's no sometimes Microsoft says, you know, this is wave one, wave two, you know, so there's no name anymore. So I'm just gonna call this the September 29 release, because that's what we have. All right.

Jordan Wilson [00:07:25]:
So anyways, what the heck is new and what are these agents? Well, they add autonomous multi step orchestration inside of office apps right now on the web. So hopefully that changes agent mode. So again, the two different agents, agent mode and office agent agent mode, embeds iterative task execution directly inside Excel in word right now. We'll see if it rolls out, to their other, office products in the near future. An office agent so I know that's kind of confusing. Actually, agent mode works inside of office's products, but office mode works inside co pilot chat, another naming thing. Come on, Microsoft. This is confusing.

Jordan Wilson [00:08:06]:
But anyways, office agent works inside of Copilot chats, and it creates, complete documents with research backed outputs. And we're gonna do I'm gonna show in, a little demo here in a couple of minutes of that one. So what can you do now? That's ultimately what it's about. Right? It's like, okay, cool. You can throw a million agents, you know, at all your users. Well, what the heck can you do today that you couldn't do yesterday? Or once, you know, everyone gets access to these agents? What can you do that you couldn't do before? Well, a lot of things. And, I've been testing out the one in Excel, and it's actually pretty good. Like, even for some of the, GBT five thinking, you know, I was trying on the new Claude four five SONNET, Gemini 2.5 pro.

Jordan Wilson [00:08:54]:
I was I was trying some data visualization things, and actually Microsoft, agent in Excel, the agent mode is handled at the best because all the other models, you know, kind of, cut off, a certain amount of data. Obviously, I was throwing, thousands and thousands of rows of Excel data at it. And, you know, luckily, Excel handled it better than the other large language models. Granted, that's not what the other front end models are necessarily good at. Right. Handling 50,000, rows of, an Excel sheet, not always the best use case, but Hey, if I'm testing it, if, if the agent mode in Excel couldn't handle that, then I'd be like, this is a problem, but it actually did really well. So here's some things that you can do. In Excel, you can do financial models, loan calculators, visual analysis and validation without expertise.

Jordan Wilson [00:09:45]:
That's the other thing. Right? You no longer have to Google for ten minutes or Google AI mode for, two minutes on how the heck do I write this formula. Now it actually just works inside of Copilot. You can actually talk to it, and it actually works. Right? Whereas before people are like, wait, why doesn't this work? Well, it does now, inside word, you can do multi file updates, structured drafting, and polished formatting through dialogue. And the, office agent, has a chat first generation of presentations and documents with live previews and checks. So let's talk about the agent mode first. Again, this is the one that works in word and Excel right now.

Jordan Wilson [00:10:29]:
So, multi step execution, that's the big thing, and it uses, the agent mode uses OpenAI's latest reasoning models to modify documents and workbooks directly. Not sure which version of GPT five it uses, or if it maybe uses the older, you know, oh three zero four mini. We're not sure it did just say it uses their latest reasoning models. So we hope that's maybe, I don't know, GPT five, GPT five thinking. Not sure. I kind of looked under the hood. I looked at the chain of, the chain of, thought summary, from agent mode in Excel. There's no identifying information, aside from it's using a reasoning model from OpenAI.

Jordan Wilson [00:11:16]:
But it does do a really good job, I thought, of planning, executing, evaluating, fixing issues, and then you can repeat and iterate until the outcomes are verified. So, it is kind of simple to just chat with the agent mode, at least in Excel. That's the only one I was able to demo. However, there were some weird issues. Right. So a couple of things that would randomly merge cells, which is not good. So I had to, like, reprompt and tell it specifically not to merge cells. That's probably a pretty big bug that Microsoft should have squashed.

Jordan Wilson [00:11:52]:
I was looking it up and there weren't a lot of people, you know, online that had been talking about this, but I saw other people had this issue. And the problem is is when you're using this thinking model and it actually does a really good job, but the problem is it's very slow. So if you do have a couple of errors, you know, you're gonna have to kind of get your prompts right. And if you're working with the same type of data or the same type of spreadsheets, you know, hopefully, you can find those, errors early on, you know, do some basic prompt engineering and get around them in the future. Alright. So that is agent mode. And then well, who can actually use it? So like I said, right now, in agent mode, this is for enterprise Copilot license and personal family subscribers via the frontier early access program, which is free and, pretty much automatic to get in. So Excel though on the web, you have to use it on the web, and it requires the Excel labs add in, and then you select agent mode.

Jordan Wilson [00:12:53]:
I have screenshot of this, so I'll walk you through it. And then it's web only for Excel and word in desktop support is coming soon. Another big bummer. So, if you are using, office agent, you will see it is in the upper left hand corner. When you choose your agents, you will see that agent there. Oh, I think I had that screenshot out of order, but that's fine. All right. So what's new in Microsoft's, new office agent.

Jordan Wilson [00:13:24]:
Alright. So that's the one that you use inside of Copilot chat. And right now, yeah, doesn't work with enterprise customers. Makes no sense. So this has the, the chat first agent inside of Microsoft three sixty five Copilot chat. It's powered this one is powered by Anthropics, Claude Sonnet four and Claude Opus 4.1. No word yet on if Claude's, anthropic Claude's just released Sonnet 4.5 will be supported or not, but at least we got, like, actual models, for Claude where we didn't get them, for, OpenAI. So this, it helps, to clarify intent, and then it also researches, previews, and generates, polished word or PowerPoint files, right in the chat.

Jordan Wilson [00:14:13]:
So, I'll play a demo of it here, shortly where we can kind of see a better, glimpse of exactly how this one works. So for this one, who can use it? How can you use it? And where, like I said, right now, it's only for personal and family, Copilot subscribers in The U S which is like no one. It just so happens that's my plan or my tier. I didn't have access. Unfortunately, I didn't have access to office agent just yet. I didn't have access to agent mode in Word, but I did have access to agent mode in Excel. So, you do, enterprise availability, is scheduled for later phases of the rollout. Like I said, I think Microsoft kinda kinda missed missed the mark on that one.

Jordan Wilson [00:15:03]:
Also English only at launch in, again, access via just the, Microsoft three sixty five Copilot chat on the web. Alright. So let's talk about agent mode in Excel and agent mode in Word and how you can access those. So for agent mode in Excel, you need to be on Excel in the web. You need to go to the add ins, kind of toolbar at the top of this, at the top of your screen, and then you need to add Excel Labs. Then you will click to use Excel Labs, agent mode, and then you are then using agent mode. So a little different, you won't click the Copilot banner. You have to actually, go into the add ins, go into Excel Labs, and then click agent mode to use agent mode.

Jordan Wilson [00:15:53]:
So little weird, because you would think, okay. Well, maybe agent mode in Excel and agent mode in Word, you use them the same way. No. You don't. Another confusing thing. Microsoft user experience. Come on. Like this should be easy.

Jordan Wilson [00:16:09]:
This is something that, you know, Chad GPT is, is great on, obviously, you know, it's, it's way more, robust. Right? Copilot, what you can do inside of Copilot. You know, it works in, like, a thousand different apps, and ChatGPT is just one. Right? So, obviously, it's a little easier. But y'all, have we not heard of simple toggles, simple drop downs, simple, right? Very easy to to navigate where to find stuff. It's it's just difficult to find these, agents right now, inside Microsoft's ecosystem. So on the word side, so the agent mode in word little different, you have to, you don't have to go into any labs or add anything. It should just be in your options toggle on the right hand side when you are, talking with Copilot, and then you, there should be a new agent mode, frontier, dropdown that you would choose.

Jordan Wilson [00:17:04]:
So, you know, you can choose researcher, you can choose analyst, you can choose designer. So at the very top, there should be a new option, for agent mode. So that's how you access the three different, agents, across the different products. So how do the agent mode products actually work? So they plan, execute, evaluate, fix issues, and repeat until verified. Office agent clarifies, on the other side, office agent clarifies intent, performs web research. That's the part that's really cool and I wanna get my hands on. I I like how the office agent, really does its work. And then it runs quality checks as well.

Jordan Wilson [00:17:47]:
But in each of these different, agentic workflows, the new agents from Microsoft users do steer goals while the agents automate procedural steps and iterations. Alright. So, podcast audience, you're gonna hear now I'm gonna play a quick, 54, demo, from Microsoft on their new office agent. So again, this is the one that for whatever reason is ultimately going to be amazing for enterprises, but it's not available to enterprises. It's only right now, this version, the office agent, is only available, to, personal and family Copilot subscribers, which makes no sense. And a quick reminder on what it can do. Well, in the Microsoft Copilot chat interface, which is what we're gonna see in this quick little demo, you can build, you know, Word docs in PowerPoints. You know, it's first gonna ask you clarifying questions.

Jordan Wilson [00:18:51]:
You can choose templates, style, multi step, web research, quality checks at every turn. So let's just go ahead and listen in our livestream audience. You should be able to watch the video here.

YouTube video [00:19:09]:
Introducing the new Office Agent in Microsoft three sixty five Copilot, the chat first way to build Office artifacts starting with Word and PowerPoint. Just describe what you need, and Office Agent gets to work. Office Agent begins by asking clarifying questions personalized to your prompt, such as key focus areas, style, presentation length, and templates. It then launches multi step web research to gather and analyze relevant info and images. You can watch as it plans how to create your presentation, executes the plan, and runs quality checks at every step, all while you steer and guide. The output is a polished deck that you can continue to iterate on with Copilot or hand off to PowerPoint and use Copilot to make edits in the app, like changing some text to a bulleted list. Try office agent today and start vibe working with Copilot. Chini.

YouTube video [00:20:04]:
Alright.

Jordan Wilson [00:20:06]:
Y'all, this is what literally everyone has wanted out of Copilot for the past two years that Copilot has been around, and it didn't come out until yesterday. And it didn't come out, for the people who actually need this, which is enterprise workers. Okay? So, again, let me quickly, walk and talk everyone through if you weren't watching the video version of this. So in the And office agent. They are inside Copilot chat, and then not having to draw from any files. So you can using the slash command, you know, you could have it read a word doc, have it read an Excel sheet, have it read a PowerPoint, and then create a new version. But in this, in this instance, in this demo, what they're doing is just having it do some research. So what it does first, it kind of creates an outline, and it goes through task clarification, which is really nice.

Jordan Wilson [00:21:10]:
Right? This one feature, I wish all large language models would do this. So, some of them kind of do versions of this. If you're doing, deep research in chat g b t, it'll usually ask you two to five clarifying questions. They're open ended though, which is fine. You have to type them out. Google Gemini deep research, gives you a plan, and then you can just approve it or edit it. So what I really like about this office, agent, it's actually really cool, at least in this demo. Right.

Jordan Wilson [00:21:44]:
So it breaks down a task clarification, and it gives you, essentially check boxes to say. So, you know, in this instance, it says presentation focus, then it gives you five different things to check, data visualization. In this case, there's two different options. It's do you want include charts for all key statistics or minimal charts focus on text, right? Like, that's great. If you're putting together a PowerPoint based on web research, having these, you know, four or five, quick clarifying questions, and then just being able to check something is ultimately going to save thirty minutes, sixty minutes, multiple hours on the backend. Right. And that's just if you were using the normal Copilot features inside of, inside of PowerPoint. Let alone if you're doing this manually, it's gonna, take way more time.

Jordan Wilson [00:22:37]:
Okay. So then what happens next after you kind of choose, from those clarifying questions, then there's you can choose a presentation length, and then there's also some built in themes. Right? So you don't have to, you know, be wondering what the heck is this thing gonna come look look like. So you can choose for presentation length, but, brief, five to seven slides, standard, eight to 12 slides, detailed, 13 to 15 slides, or comprehensive, 16 to 20 slides. Y'all, this is what everyone's wanted out of Copilot for forever. Right? Go do a bunch of research, you know, ask me a couple of questions, and give me a decent looking PowerPoint. Right? This is your, you you know, management consulting one zero one. This is what management consultants do.

Jordan Wilson [00:23:19]:
Alright? So then after you choose the presentation length, after you choose your theme, you submit. And then at this point, this is when it goes out, the office agent, and it starts to do, web crawl here. So it does multiple web searches, right, just like a deep research would. And again, this version here is powered by, Anthropics Claude's models, which are great at agentic research and just agentic work in general. So it goes through, it does a couple of web crawls, a couple, follow ups looking for more information. In this instance, it's looking at images online, and then it's planning the slides. Right? So at each of these, iterations, it looks like you can stop and, you know, edit things, but it's starting to plan all of these slides. It, on the right hand side, it's starting to show a preview.

Jordan Wilson [00:24:17]:
You can go in kind of read the preview. You can also modify it in natural language. So in the demo video, it says I'll modify slides, or sorry. The, the prompt was removed the image in slide six and turn the bullets into a visual. Great natural language prompt. So as you see, your PowerPoint slides being built, if something's not right, you can say simple, Hey, slide blank, get rid of this, change this. And it seemingly does it fairly well. And then at the end, it you I mean, has a decent looking PowerPoint.

Jordan Wilson [00:24:52]:
Right? So, this, you know, from an AI perspective, it's it's it's not, you know, at the level of, like, a gamma, or something like that that I think has very clean visuals when it comes to, you know, AI presentations. But again, at least on this demo, fantastic. And I'm very bullish on this, but what the heck? Why Microsoft? Aren't you giving this to people that probably actually want to use it? Right. You're gonna force all these enterprises, to, you know, get a personal or, family, copilot plan just to start using it. Alright. So let's keep this thing going, and talk about what specific models each one knew. So, each agent uses. So like I said, agent mode, uses OpenAI's latest reasoning models.

Jordan Wilson [00:25:44]:
And then Office Agent, the one I just showed a demo of, which doesn't work in Office, it works in Copilot chat, that uses Claude's model. But I do think Microsoft is smart with this multi model strategy, right, using the right model for the right reasons, which is actually pretty good on Microsoft considering they have an equity stake in OpenAI, but they don't, in Anthropic. So, in one case, they are actually paying, ultimately, to use a model that they say is better than the one that they have equity in. So, you know, hats off to Microsoft for, presumably doing the right thing. You know, I'm sure their internal testing on this was rigorous. Right? Imagine having to be the one to make that call. Hey. You know this model that we have equity stake in? Well, we're not gonna use it, for this agent.

Jordan Wilson [00:26:34]:
We're gonna pay probably a ton of money, to use, a model, from anthropic because it gives better performance. So what's this multi step orchestration and how is this really different? Well, this breaks high level goals into subtasks. Right? So, these agents, in all the demo videos, and we'll be sharing those, in our newsletter today. If you wanna go watch those, it does break it down into multiple steps. So in my, demoing, of the agent mode in Excel, it actually did a really good job of that. And I was fairly impressed, and Excel, creates it. Can it can create, sheets, formulas, charts in summaries with validation driven loops. And then in word, the word one, I wasn't super impressed with if I'm being honest, the agent mode in word.

Jordan Wilson [00:27:25]:
I don't know. I would probably much rather, use Chat GPT canvas mode, Google Gemini canvas mode, Claude artifacts. Probably not Claude artifacts. I'd probably use Chat GPT, canvas mode or Gemini canvas mode. Again, Microsoft pushing this whole, like, vibe writing, vibe working. No. Again, I I probably wouldn't use, the agent mode in word over the long run. The only, instances or the huge value, obviously, again, if you are a heavy Microsoft three sixty five organization and you have docs that are constantly changing, you can obviously use the slash command, in word reference, multiple docs and create a new doc, pulling key information from your other files that are up to date across your organization.

Jordan Wilson [00:28:11]:
That's you know, you can't overlook the value add of that. But aside from that, agent mode in Word, agent mode in Excel, loving it. Office agent. I will love it once everyone gets access, cause it's going to be super confusing. So let me just tear this apart real quick. Microsoft kind of out of nowhere, just going with this vibe working. Right. And getting back to the heart of this episode, right? Is this the new way that millions of people inside windows are going to work? Right? This new agent mode and office, office agent, or is it just fancy marketing? And this just, I don't know.

Jordan Wilson [00:29:01]:
I don't know what it was that, you know, it was really grinding my gears about this one. It's, it's not nearly as bad as when apple rebranded apple intelligence, you know, or they rebranded artificial intelligence and try to call it apple intelligence, but I don't know Microsoft Vibe working. Right? The literal headline, of their agent mode, release on their website said Vibe working, introducing agent mode and office agent in Microsoft three sixty five copilot. I get what they're going for. Right? Like saying, hey. You know, now, especially in, Office agent, right, where you can just create PowerPoints by, you know, natural language prompts and, you know, being able to pull information from across your other Microsoft documents. So I get the concept, but it's not there. It's not there because I know the Copilot ecosystem is problematic.

Jordan Wilson [00:29:55]:
I've talked to, and this is not an exaggeration. I've been doing this for a long time. I've talked to hundreds of business leaders that are like, yeah, we pay for Microsoft CoPilot licenses, but we can't use them because no one knows how. Right? It's almost like Copilot is in too many places, but the places where you wanna use Copilot, no one understands it. No one understands the user permissions. Right? Getting people, the right access. People are not sure. Oh, should I be using, you you know, Copilot in, you you know, MM365 chat? Should I be using it, in Teams? Should I be using it inside Microsoft Word? It's almost in way too many places, and, you know, file permissions, admin, issues.

Jordan Wilson [00:30:41]:
It's it's very confusing. So many large I'm talking Fortune 500 companies, that I've talked to have Microsoft three sixty five, and they're still instead using ChatGPT or Gemini because it's confusing. So Microsoft, instead of trying to create a category, right, of, like, we are the vibe working platform, you know, how about you just make co pilot work a little bit better? Right. And this, and this rollout is exactly what I'm talking about. Why, why would you roll out this office agent? That is, like I've said, I'd hate to beat the dead horse, but sorry, dead horse. I'm going to beat you one more time. This is what people have been wanting Microsoft, right? An easy way to create an actual great, looking, presentation, with an actual agentic experience, right? This multi step orchestration, working with all of your files, doing multiple steps of research, being able to iterate and, and observe along the way. This is what everyone's wanted.

Jordan Wilson [00:31:51]:
Yet. Good luck. Anyone getting access. Right. Because at least according to their release, it's not available to the people that I think actually need it and could push this thing forward. Right? And people, you know, are, and I talk to people, you know, kind of off the record at Microsoft all the time, and many of them are aware of this. And you wonder why more people aren't raving about Copilot. They should be, but this is one of the reasons, right? The rollout is jagged.

Jordan Wilson [00:32:21]:
Permissions are confusing.

YouTube video [00:32:25]:
And the

Jordan Wilson [00:32:25]:
go to market is non existent. This is not a strategy. This is a strategy for confusion. And for when it's finally available for everyone else, they're gonna be like, wait, this is old. No, it's not right. When, when, thousands or tens of thousands or hundreds of thousands of companies get access to this, whether that's in a couple of weeks or a couple of months, they're gonna be like, wait, this already rolled out and it didn't work for us. So why would we look at it again? You don't get five shots at a grand reopening. You get one.

Jordan Wilson [00:32:59]:
So a little bit more on multi step orchestration, and then we're going to wrap today's show up. So this is where you just state desired outcome. And I do think it's worth talking about again, because this is one of the big differences between, these new agents and what was already available inside of copilot or what was already available inside of copilot studio. And this is where Microsoft was, you know, really trying to go hard with this vibe, vibe working, thing. But this is where you just state, you know, end users state their desired outcome, and hopefully, Copilot handles the complex steps conversationally and iteratively. Hopefully, it speeds up. Like I said, I was happy with the results that I was doing in agent mode in Excel, but sometimes like when it merged cells randomly, I had to wait like five minutes, for it to unmerge those cells. Right.

Jordan Wilson [00:33:53]:
So, the iterative nature is great. The multi step orchestration is great, except if it's really slow. Right? So that's one huge downside. And this does adapt these kind of vibe coding principles to mainstream productivity for non experts. Right? So anyone, if you haven't vibe coded, right, that's where all these platforms, I mean, there's dozens, you know, Cursor, Lovable, Bolt. Right? Google has some great ones where you just talk to and you're like, hey. Build me an app that does a, b, and c. And then it spits out the first version.

Jordan Wilson [00:34:26]:
You're like, oh, yeah. This is good. But let's, you know, make it make it cleaner. Let's let's add in a database, etcetera. Right? So you're just literally talking something into existence. So that's what Microsoft is going for here with this, new, multi step orchestration in, agent mode in office agent, but I just think it's gonna fall short a little bit. Limitations. What about that? Well, it's frontier only preview, so you have to be part of the frontier, program.

Jordan Wilson [00:34:56]:
It's only web first, which is another, downside. I'm guessing that might be one of the reasons why it's not, some of these things aren't rolling out to the full enterprise, customers right now. Office agents is limited to US and English during the initial preview phase, and the Excel agent mode, although it did really good, it it actually benchmarked really, really well. So I have this benchmark here. This is from the spreadsheet bench accuracy results from Microsoft's website. It actually beat out a lot of the other, agents, on the spreadsheet bench accuracy results, with a 57, point two. So the the downfall here is, well, still not as good as a human. A human scores about 71%, on the spreadsheet bench, accuracy.

Jordan Wilson [00:35:42]:
Although, the Copilot in Excel agent mode did much better than all of the other agents, on this benchmark. So, it's a step in the right direction, but it's not ready to necessarily replace, your data analyst. But I do think, especially for those that are, you know, either number one, creating, a ton of PowerPoints, which is a lot of people, or spending a ton of time in Microsoft Excel, which I know is also a ton of people. That's why you have to pay attention, to the unfortunately tiered and delayed rollout because once this becomes available to everyone, I do think it's worth paying attention to. And you might be saying like, well, couldn't you do this before? No. Previous Copilot was more of a single turn assistance without any autonomous multi step orchestration, and users had to manually either re prompt from scratch, because there was no iterative, and evaluative and fix, loops that existed before. But this new release, does enable that proactive execution and verification inside of office apps. So, yeah, that's another big difference.

Jordan Wilson [00:36:49]:
You know, maybe you're using, you know, Copilot and Excel before. What's the difference? Well, it breaks things down, into this multiple step orchestration, but also the iterative nature. Right. So you don't, whereas before to get better results, you just kinda have to reprompt from scratch. And now you can just say like, yo, change this one thing. Right? In the example, of office agent in the demo, it was, you know, Hey, get rid of the picture in slide six and change the bullet points to something more visual. Right? Being able to, take just one step out of the multi step, orchestration that the agent did and just change that one thing versus having to re prompt the whole thing from scratch. So obviously, that's a huge update.

Jordan Wilson [00:37:32]:
And then you're like, okay, Copilot studio. Didn't we have this? No. So, although you could probably, build a lot of not a lot, but you could probably build, some of the things that you might use agent mode in Excel, agent mode in Word, or Office Agent four. You could probably build some of these, capabilities, in Microsoft copilot studio. So Microsoft copilot studio for people that don't know, it's a low code, sometimes no code platform, for building custom organizational agents that then you can use throughout your organization. The biggest difference here, well, agent mode and office agent, they're prebuilt. Their end user features requiring no development. Copilot studio, like I said, you gotta go in and build it.

Jordan Wilson [00:38:15]:
So agent mode is already pre built pre configured it's for using co pilot studio is for building. So if you confused, that's kind of the big differentiator. All right. So that's a wrap, but I want to leave you with a couple use cases that you can try today. All right. So agent mode in Excel, you can run a full sales analysis with charts and validated insights from a single prompt. Alright. Agent mode in word.

Jordan Wilson [00:38:45]:
You can update a status report. You can pull data from multiple different files, by using the backslash command and referencing those different files. You can do the same thing, obviously, in Excel, Excel's agent mode and then office agent. So using that inside of chat, once it rolls out to enterprises will be fantastic. So this is, the ability to generate a market trends, PowerPoint with live multi step validated web research and previewing your slides before, making them. Alright. I hope this was helpful y'all. I had personally, I had so many questions, about these new Microsoft agents and if they were more marketing or the future of work.

Jordan Wilson [00:39:32]:
And, well, my hot take is little of both. I think this rollout could have been much better by Microsoft. However, in the long run, I think these might be the agents that actually get people using and paying more attention, to Microsoft copilot where they maybe hadn't as much before. So I hope this show was helpful. I hope maybe you got some of your questions answered, but if this was helpful, please consider reposting this, on LinkedIn. We always send that to, the live stream to our, to our LinkedIn account. If you're listening on the podcast, appreciate your support. If you could, subscribe, follow the channel and leave us a rating.

Jordan Wilson [00:40:11]:
And then when you're done, please go to your everydayai.com. Sign up for that free daily newsletter, and then we'll see you back tomorrow and every day for more everyday AI. Thanks, y'all.

Gain Extra Insights With Our Newsletter

Sign up for our newsletter to get more in-depth content on AI