“Boy, that escalated quickly.” Time to recycle the image from my earlier take, AI Man: The Legend of Prompt Burgundy.

I knew my previous newsletter issue would get the attention of many people in the Microsoft ecosystem. Because I wrote about something you don’t hear too often in these circles. It’s not common at all for someone to openly point out that according to the leaked numbers, Microsoft 365 Copilot as a product has not been a roaring success but rather a commercial failure:

The comments on LinkedIn were understandably skewed towards the “I don’t agree with this at all” end of the spectrum. Many folks who follow me on that network are working as Microsoft partners, trying their best to succeed with the products and services that have been given to them. It’s important to keep in mind that I don’t claim these people have failed in their jobs. Neither did I say that no customer could get a positive ROI from M365 Copilot rollout. Yet this was what many seemed to be reading between the lines when I pointed out the poor MS license sales figures.

What about different types of audiences? How would people who do not have a commercial interest in MS license sales (which often is indirect, BTW) react to the words I wrote in this newsletter? I haven’t been sharing my earlier posts to the Hacker News discussion forum, yet this time I decided to make an exception. I figured this was a much more broadly interesting topic than my niche Power Platform / low-code / licensing rants.

There was again a lot of interest in the article. This can be seen in the new subscriber peak for Perspectives on Power Platform below. So, if anyone found this newsletter via HN, then I want to wish you welcome into this bizarre yet intriguing world of Microsoft business apps and the business of software!👋

The 169 comments on HN had a different tone than the 149 on LinkedIn. Now, in general Microsoft still gets a lot of (IMHO) unwarranted “blame by default” in tech forums where they are mentioned, coming from the pro-dev and open-source oriented audience. I can understand the reasons quite well, though, given the company’s history and how they’ve sometimes abused their market position.

Which side is right then? Neither one, of course. There’s truth in a lot of what both the partners and the hackers are saying. Since MS happens to be one of the biggest corporations in the world, the market reality is a mixture of fanboys, haters, and everything in between. They operate on a scale that is nothing like a typical software company building cool apps. Microsoft operates more like a nation-state covering the entire spectrum of people who use computing devices - or are impacted by how such devices operate.

Looking at annual revenue, Microsoft makes as much money as a nation like Sweden does in estimated tax revenue collected. Twice the amount of my home country, Finland. The number of paying customers for Microsoft 365 is close to the total population of the EU. Using a revenue multiplier often cited by MS, their partner ecosystem generates annual revenue equivalent to the GDP of France (7th largest economy in the world).

Are citizens allowed to criticize the actions of political leaders running such countries? Hell yeah! Do some of the investments made by nation state governments fail to meet the publicly stated success criteria, despite the administration claiming otherwise? All the time. Hasn’t it been crucial for their chosen form of government (democracy) that there’s room for expressing thoughts that differ from the official narrative of leaders? Today, a lot of us are seeing the value of this right that we took for granted before becoming something that needs to be fought for once again.

Sure, Microsoft isn’t a democracy. It’s a business. I believe the best way for people who aren’t major shareholders of such businesses to influence the way they operate within society is by keeping them honest. That’s one of the key reasons why I bother writing about these perspectives of mine online. I’m always delighted that someone else bothers to spend time reading them, too. Including those who don’t agree with everything I say.

First of all, it’s not a small number. Second, we can’t know exactly where it comes from and how it is calculated. That’s because Microsoft isn’t keen on sharing the official customer numbers. While I was chatting with Claude to create an interactive “quiz” to evaluate the commercial success of M365 Copilot, the LLM spit out something related to this that I thought was pure gold. So, I wanted to share the AI insights.

Here are some common red flags in Microsoft’s communication when their products are not performing as expected:

Reading that table, I realized large language models may not be the right tool for all business problems (especially anything dealing with numbers). Yet when it comes to analyzing the language used by corporations, I consider this the perfect job for AI chatbots. After all, the foundational models have been trained with all the PR text found in the public internet. If anyone can spot a communication pattern, an LLM surely can.

On the human intelligence front, some readers criticized my conclusions about commercial failure being too shortsighted. Not giving enough time to Microsoft 365 Copilot get adopted inside slow-moving enterprise organizations. That’s fair, and I certainly have faced this myself repeatedly when trying to evangelize Power Platform as a citizen developer platform that businesses should pay premium licenses for. 2 years is not a long time in the business IT world.

Now, I just wish people would keep this in mind when breathlessly posting how “this latest AI game-changer just killed established software X and you need to get on board RIGHT NOW” on LinkedIn. Both things can’t be true simultaneously. I find it hard to understand why we should adopt entirely different standards for evaluating the credibility of claims, depending on whether the person is A) repeating and enforcing the narrative from a tech vendor, or B) questioning whether said narrative is supported by facts from independent parties.

The reason I think two years is enough to judge M365 Copilot’s commercial trajectory is because the life for individual software products is getting shorter every day. We’re way past the old days when waiting for a V3 of a Microsoft product was the smart way to do things. In those days, you often also had years in between each version developed and launched. Today, no versions exist anymore, and the products get reimagined in an instant by MS product marketing if the targets aren’t met.

Speaking of targets, let’s look at another anonymous source sharing their experiences of Microsoft’s Copilot business strategy and its outcomes. In the Pivot to AI YouTube channel that shares daily AI parody content, the video about “Microsoft: forget the IT dept, use Copilot anyway!” had an interesting comment from a person working for a large publishing company in the Nordics:

Allegedly, a VP level representative of Microsoft had told the customer that the only thing she is being measured on for Q4 (presumably of FY24, meaning Apr-Jun) was sales of Copilot. When the customer wanted to talk about purchasing non-Copilot licenses, they weren’t of any interest for this VP to talk with anymore.

I also love the note about Azure storage discount negotiations. How MS sales reps are so desperately trying to show they’ve managed to do what Redmond demands from them. So, despite the customer’s rejection, why not try to sneak Copilot back into the contract when they aren’t looking? Obviously, they just don’t understand what’s good for them!

Here’s the thing: if businesses indeed are adopting the Copilot products according to the planned growth curve, why is Microsoft acting so desperately? While the above YouTube comment is just an individual example, when talking with folks who do licensing contract negotiations for a living, this appears to be a recurring pattern with Microsoft representatives. It’s Copilot or GTFO. Which sounds to me more like a sales organization under immense stress to meet impossible targets, rather than a confident software vendor that thinks the Copilot adoption rate is evolving at its natural pace, according to plans.

There’s a reason why Microsoft is still around, 50 years later, and with a bigger market cap than ever before. I don’t believe that the struggles with selling paid Copilot seats will slow them down in the long term. What we’re witnessing is just one of the episodes in the business of software that MS knows darn well how to play. Besides, they’re hardly the only Big Tech company struggling to turn hype into revenue right now.

What I believe will happen is that when a sustainable business model is eventually found (after the AI bubble has first been allowed to burst, most likely), no vendor will address the mistakes made with the first attempts of AI products we see out there now. When it comes to episodes of the tech business, there’s no expectation of a coherent storyline continuing across episodes. As Satya himself famously said, the industry does not respect tradition, only innovation.

Innovation will continue until revenue improves, that is the logical expectation. If I were working at Microsoft and I’d have to figure out an innovative way out of this pickle, without directly admitting any failure, here’s how I’d do it.

[Begin presentation]

First, we can’t start from scratch again. Yes, that 8 million user base may only be 2% of the total addressable Microsoft 365 subscriber market. They may not be madly in love with the products they’ve signed up for, but still, we have a contract with them! You don’t just walk away from M365 Copilot subscription the next month, unlike with SaaS AI plans sold to individuals by new tech startups. Whatever we do as MS, making the most of this early adopter audience is critical.

Second, we’ve already basically admitted that the Office in-app Copilot features aren’t that valuable. When giving away Copilot Chat for free, first on the M365 home page and now inside apps like Word, Excel, and PowerPoint, we’ve muddied the water even worse than it was before. Heck, even Microsoft MVPs are now spreading the word out that “Copilot is free inside M365”, thanks to how our tech evangelism team is presenting it. Deciphering what features require a paid subscription from the documentation is impossible without resorting to some AI assistance.

Okay, so we did leverage the sales argument for the paid Copilot features just recently when launching the M365 Premium for individuals, allowing them to bring their own Copilot to work. But since we’ve already forced the consumers into paying extra for Copilot a while ago, they’re not going to hate us any more if this turns out to be a non-feature very shortly.

Here’s my plan: let’s reimagine the Microsoft 365 Copilot $30pupm subscription to be called Microsoft Agent 365 instead. Same SKU, same existing customers, same contract terms, same price, just a different positioning story. Brilliant!!!🎉🎉

Copilot was a nice concept, and even a pretty slick brand name when originally rolled out for GitHub. But let’s face it: we’ve messed it up big time. As Satya recently admitted, no one can keep up with the different Copilots anymore. People today are as passionate about using Copilot than they are about Windows. But not like in the Win95 “Start Me Up” days, but more as a thing they’d replace with a Mac if they had the power and money to do so. As we know, people aren’t choosing Copilot, they’ve chosen ChatGPT instead. Among enterprise customers, 90% of users go to ChatGPT and Copilot is on the fourth place after Gemini and Claude. Ouch!

Agents, on the other hand, are something that everyone else is doing the marketing for. Just look at OpenAI. A few weeks after we renamed M365 agent builder to Copilot Studio lite experience, Sam and his crew launched Agent Builder in ChatGPT. Doh! While we keep renaming things to not be agents, like with Power Virtual Agent rebranded as Copilot Studio, the industry is going the opposite direction. Copilot is in danger of becoming the Clippy 2.0 that Marc Benioff mocked it to be a year ago.

It’s time for Microsoft to take back control of the narrative! Naturally, Copilot will not be removed from the millions of places where we have managed to inject it to during the past couple of years. Even renaming it would be a herculean effort at this stage. Instead of hoping customers would pay money for something that says “Copilot”, we’ll flip the story around. Agents are where the price tag is going to be attached to from now on.

We’ve started the preparation work for this already. First, we removed the need to pay for advanced features in Copilot Studio agents if you’ve got the Microsoft 365 Copilot license. Agent flows and AI tools are now bundled in, no Copilot credits consumed. Second, we’ve rolled up the individual Copilot for Sales, Copilot for Service and Copilot for Finance licenses into M365 Copilot for no extra fee.

We’ve branded even basic NLP features used via an app GUI as “agents” for the past year now. We introduced a feature in SharePoint that resulted in millions of new agents being created (with a 0.1% weekly active usage rate in Aug 2025). This means we have inflated the numbers of individual agents that we can count in our big numbers to publicize at a suitable moment, like Ignite 2025. We’ve got a team working on the new enterprise scale offering that we may or may not call Microsoft Agent 365:

It’ll be amazing! By framing the early adopters of premium AI products into having embraced “enterprise-grade agentic AI” with their choice of buying Microsoft 365 Copilot licenses (now M365 Agent), none of them need to admit that blindly buying the Copilot seats based on video demos alone wasn’t a smart move. Instead, we can include them in day-one marketing videos as testimonials from existing customers who have bought the product that we now call a different product. Instant success and validation!

Taken from our Power Platform playbook, the key is in finding compelling arguments that require companies to purchase wall-to-wall licenses for all users to gain the enterprise-grade benefits of agent control. The current limited pilots that customers have made by purchasing a handful of M365 Copilot licenses will not organically grow into org-wide contracts. Because that would just give them more of the Copilot things we’ve claimed to build yet not quite managed to meet the expectations for two years now.

No, our true commercial success will not come via inventing an Office subscription that costs twice as much as before. Instead, we must position our offering as the most credible enterprise platform to manage any AI agents at work. The more comprehensive we make it sound, by leveraging our existing footprint in collaboration, security, development platforms, the harder it will be for anyone (read: OpenAI) to challenge us. Our sales decks for Agent 365 will be dynamically constructed from our massive library of cloud pitches that we fed to our MAI-1-BUY in-house foundation model.

And that, my friends, is how we’ll transform our V1 Copilot failures into a V2 Agentic success. By reimagining, renaming, and re-licensing. The three core pillars of our commercial strategy.

[Mic drop / Teams call disconnect]

As mentioned, people have diverging opinions on the state and direction of the Copilot business. So, I want to invite you to vote in the following poll that gathers some data about the sentiment around M365 Copilot’s commercial model:

Please do feel free to leave comments on this newsletter issue. Or reply to me privately via email if you prefer to not make any publicly visible statements. A big part of how I produce these Perspectives of mine is based on direct discussions and reading what others in the ecosystem share. I then try to find a way to distil all of that into a newsletter that helps the community make smart choices around the Microsoft technology and business that unites us.