Microsoft's Evolving AI Strategy: New Clues from Build

Summary

---

Podcast Summary: Microsoft's Evolving AI Strategy – New Clues from Build

Podcast: The Directions on Microsoft Briefing Podcast
Host: Mary Jo Foley (A)
Guest: David Berry (B), Directions Analyst
Date: June 11, 2026

---

Overview

This episode explores Microsoft’s AI announcements and overarching strategy as unveiled at the Build 2026 conference—a flagship event for developers and enterprise customers. Mary Jo Foley and analyst David Berry break down the flurry of AI agent launches, hardware advancements, and what it all means for enterprises. The discussion covers Microsoft’s AI model proliferation, the implications for OpenAI partnerships, agent governance and licensing, potential platform lock-in, and how enterprises should approach AI investments in an era of rapid change and increasing costs.

---

Key Discussion Points & Insights

1. Microsoft's AI Model Strategy & Relationship with OpenAI

• Microsoft is accelerating the development of its own AI models without breaking its essential ties to OpenAI.
• “I actually don't think it changes much about their relationship with OpenAI for right now. I think that the big shift already happened when they agreed to see other companies. But they both still need each other, right?” — David (02:05)
• The shift toward a multi-vendor, multi-model ecosystem (with relationships like the recent Anthropic partnership) is now central to Microsoft's AI messaging.

2. Model Choice and Multimodal Workflows

• Microsoft emphasizes “model choice” as a point of pride, giving customers flexibility in choosing GPT, Claude, or Microsoft’s own models within tools like Copilot.
• New workflows allow for combining multiple models to improve accuracy and reliability (e.g., using GPT for drafting and Claude for critique).
• “Model choice is here to stay—people want the right tool for the job. But I think... they're going to have to start paying for the privilege.” — David (03:10)
• Microsoft’s own models may become defaults or the lowest-cost option, with third-party or premium models commanding an upcharge.
• “By introducing their own models at a lower price point, they can give companies the opportunity to use them... Microsoft may replace the default models with their own and then upcharge for the use of the Premier Frontier Models.” —David (05:21)

3. Agentic Platforms and New Always-On Agents (“Scout”)

[07:06]

• Enterprise trust in new AI agents like “Scout” (running on OpenClaw) is still developing, mainly due to security and governance concerns.
• Microsoft's new agents will have unique identities, mirroring user management models (with email, Teams accounts, etc.).
• “The idea of an autonomous agent wandering through their data is probably unsettling. Governance hasn't even caught up with what we have right now.” — David (07:32)
• Licensing model for such agents likely to be user-based, perhaps with cost-plus-usage pricing.

4. AI Hardware and Local Model Execution

[10:44]

• Microsoft announced high-end ARM-based hardware (e.g., Surface RTX, RTX Spark dev box) to enable local running of AI models.
• Business case mainly targets organizations seeking to reduce cloud GPU costs through local development/inference.
• “It’s not the developers they have to sell, it’s the bosses. The dev box is targeting local model development—not just your typical developer. This is a smaller target market, but a viable one for companies looking to avoid cloud GPU expenses.” — David (10:56)
• Hardware cycles in enterprises (usually 2-3 years) add complexity to widespread hardware adoption.

5. The "Copilot Super App": Vision and Skepticism

[14:20]

• Despite rumors, Microsoft didn’t unveil the “Copilot super app” at Build, but CEO Satya Nadella teased a summer announcement.
• The panel debates whether a universal Copilot interface (for both consumers and enterprises) is feasible or desirable.
• “Unless we’re talking like Star Trek’s computer, I’m skeptical of the super app concept...They’d have to invert the premise from embedding Copilot where you work, to making the Copilot app be where you work.” — David (14:20, 16:22)
• Mary Jo highlights confusion due to proliferation of Copilots and Microsoft’s likely need to consolidate the experience.

6. Microsoft’s True AI Play: ROI, Lock-In, and Strategy

[18:19, 19:15]

• ROI is a problematic lens for evaluating AI investments; Microsoft is playing for “platform dominance,” not near-term profitability.
• “Microsoft is willing to sacrifice short-term gains from Copilot and other AI initiatives to position themselves for the longer-term shot at platform dominance. It’s a playbook they’ve used before.” — David (19:22)
• The source of enterprise lock-in is shifting from models (which are commoditizing rapidly) to data, workflows, and integrated security/governance features—areas where Microsoft can add unique value.
• “Models are almost a commodity now... but Microsoft’s value comes from integrating with M365, providing business context, security and governance.” — David (21:14)

7. Enterprise Advice: Best-of-Breed or Microsoft Platform?

[24:01]

• The guest favors a nuanced “it depends” approach based on risk tolerance, business objectives, and current investment in Microsoft technology.
• “If you’re a Microsoft shop, it may make sense technically to continue down that path, but your relationship has changed and costs are rising. If you're in the business of AI, building products and services, best of breed may make more sense. It depends.” — David (24:09)

---

Notable Quotes & Memorable Moments

• "It sort of feels like we went from crawl to run and skipped walk." — David Berry, on pace of agentic AI rollout (07:06)
• "The idea of an autonomous agent wandering through their data is probably unsettling." — David Berry (07:32)
• "ROI is the wrong lens... Microsoft is willing to sacrifice short-term gains... for the longer-term shot at platform dominance." — David Berry (19:22)
• "Are we really just going to get something that's mediocre across the board?" — David Berry, on the risk of a jack-of-all-trades “super app” (17:57)
• "I don't think most people are going to want to sit around and get into the nuances of which model is best for creating a PowerPoint. So, yeah, they're going to turn to their agents and get the task done." — David Berry (04:44)

---

Important Timestamps

• 02:05 – Microsoft’s continued but shifting relationship with OpenAI.
• 03:10 – Multimodal workflows and model choice as differentiators.
• 05:21 – Cost strategies around Microsoft’s in-house models.
• 07:06 – Trust and governance issues with autonomous AI agents.
• 08:46 – Licensing implications for always-on agents.
• 10:44 – Hardware innovation and its appeal for local AI processing.
• 14:20 – Speculating on the Copilot “super app.”
• 19:15 – Microsoft’s long-term strategic approach (beyond ROI).
• 21:14 – Where lock-in is shifting in Microsoft’s AI stack.
• 24:09 – Concrete advice to enterprises about AI stack choices.

---

Guidance for Enterprises

• Evaluate risk and prepare for evolving costs: Microsoft is shifting to usage-based pricing and may drive platform lock-in deeper into workflows and data integration, not just the AI models.
• Platform approach vs. best of breed: Use a platform approach (like Microsoft’s) if you prioritize simplicity and integration; consider best-of-breed if AI development is core to your business.
• Stay flexible and negotiate: With frequent product changes and rising prices, IT leaders should prepare to renegotiate terms and track the evolution of Microsoft’s AI licensing and platform strategies.

---

Tone

In-depth, candid, and pragmatic—Mary Jo and David provide both high-level perspectives and granular, actionable advice, echoing the practical, enterprise-focused ethos of the Directions on Microsoft Briefing Podcast.

---

Transcript

[A]

Last week, Microsoft held Build, one of its two biggest annual conferences. There are all kinds of announcements from an Android based device platform for agents to seven new or updated Microsoft developed AI models. Today, after taking a week to digest all the news, we're going to talk about what we learned from Build and go deeper on some of the products and services announced there that we think will matter to enterprises. Welcome to the Directions on Microsoft Briefing podcast. I'm Mary Jo Foley, the editor here at Directions on Directions. I'm your host for the series of podcasts for those interested in the Microsoft enterprise IT ecosystem. My guest today is Directions Analyst David Berry. David has more than two decades of architecture and delivery experience and positions with Avanade, Booz, Allen Hamilton and emc. And just before he joined Directions at Avanade, he led global and regional engineering teams and advised on technology strategy and adoption. Welcome David, thanks for joining me on the episode today. It's been too long since you've been here.

[B]

It has. Mary Jo, thanks very much for having me back. I've missed it and look forward to today's episode.

[A]

I'm so glad you want to talk about Bill, because I always like digesting and dissecting these kinds of events.

[B]

Yes.

[A]

Let's start out with some of the many agent and AI model announcements that Microsoft made it built. As we've all noticed, they're really putting the pedal to the metal to develop and roll out all their own AI models lately. So curious what you think here. What does this say about its strategy and relationship with OpenAI?

[B]

Well, I actually don't think it changes much about their relationship with OpenAI for right now. I think that the big shift already happened when they agreed to see other companies. But they both still need each other, right?

[A]

True.

[B]

Yeah. The break in this partnership with Anthropic, though, I think was the beginning of Microsoft's pivot through to a multi vendor, multi model provider, which seems to be part of their bigger plan.

[A]

Yeah, I want to ask you more about that because I think even though they're doing all these different things on their own with their AI models, they seem to still talk a lot at Microsoft about model choice. Take that as a point of pride. So I'm curious if you think they're going to continue to hold that stance as their own models evolve and proliferate. Do you think customers are still going to have the option to use like GPT or cloud inside Microsoft 365 copilot or GitHub copilot? What do you think is going to

[B]

happen Well, I think model choice is here to say sort of people want the right tool for the job. But I think what's going to happen is they're going to have to start paying for the privilege. CLAUDE and GPT will continue to be available barring any falling out with Microsoft. Then Microsoft is introducing its own models, so they'll be part of the mix. There's even a reference architecture for companies to add their own models into copilot. So I really don't see choice going away. The other thing is that Microsoft is starting to focus on some multimodal workflows. What I mean by that as an example is if you take a look at Microsoft Researcher, that agent can now use both GPT and CLAUDE models to generate a response. For example, in critique mode, which is now the default for Researcher, it uses one model to generate the response and the other to check the response. And then Council uses both models to generate responses and a third judge model to assess the results. These are generally going to increase the accuracy and reliability overall. Microsoft, I think, is liking the direction of doing this. Yeah, but I think what's really going to happen is choosing the model is not the choice that we make. Um, you know, as I said, people want the right tool for the job and I don't think most people are going to want to sit around and get into the nuances of which model is best for creating a PowerPoint. So, yeah, they're going to turn to their agents and get the task done. And I think the agent is going to be where, you know, people are going to be making the choices.

[A]

Interesting. Okay. I also think, I think you brought this up on one of our edit calls that we have internally. I think what they're, where they're going is their models may become like the default. If you don't pick a model, you get the Microsoft model and, or may become the cheapest option as a way to kind of drive people that way.

[B]

Right, right, exactly. So they made a big deal about the models when they introduced them about being cost effective. And I think we're seeing a lot of cost pressure come to the forefront now, not just with Microsoft, but across the board. And by introducing their, their own models at a lower price point, they can give companies the opportunity to use them. But I think as we also watch what Microsoft is doing with their pricing models and this shift that they're signaling to cost plus usage. What we actually may see is Microsoft replace the default models with their own and then upcharge for the use of the Premier Frontier Models.

[A]

Yep, yep. I think that's where we're going. That makes sense. Okay, so we're talking about agents and all things agentic. Of course. That was a huge theme at build. They even took the wraps off something I was surprised about. Their, their thing they call Scout, which is a personal agent that Microsoft is building on top of openclaw. So I've been following this since it was basically the germ of an idea 60 days ago and now they're ready to throw it out there into the Frontier testing program. I. I am stunned how quick happen. So I'm curious if you think Scout and these other kinds of new always on agents that Microsoft is building may change the perceptions enterprises have about things like Open Claw not being ready for the enterprise in terms of security and governance and probably other agents as they're announced.

[B]

Yeah, it was, it was interesting to see the demo of Scout there. I think there's a lot going on with Scout even before we get to openclaw. If you're talking about trust in the enterprise, there's a much longer road to hoe. It sort of feels like we went from crawl to run and skipped.

[A]

Yeah, totally. Oh my gosh. Yeah, especially with this.

[B]

The way I look at it is most enterprises are still dealing with, you know, data cleanup and security for just the AI capabilities that we have already. So the idea of an autonomous agent wandering through their data is probably unsettling.

[A]

Yes.

[B]

Yeah. Governance hasn't even caught up with what we have right now, let alone, you know, get ready for autonomous agents. Having its own identity raises questions about onboarding an agent. You know, what the exit process looks like, what group memberships does it have, what permissions, what policies need to be added. There's a whole lot to it before, in my opinion, that it can really become viable. I mean, everything you have to do for a user, you have to do for Scout and these autonomous agents basically.

[A]

Right, right. I know they've been saying they were going to do that, like make come out with a category of agents where they had their own email account, their own team's presence and all that. I felt like maybe we were further away from happening, but it's like no Scout is going to be the first thing that has that. And I, it really makes me question like, what's the licensing model going to look like for these always on agents? Are you going to license them like a person? I guess so. Right?

[B]

Yeah. I mean that's, that's my take on it. I mean, they. Microsoft execs have been messaging this over the last year or two that, you know, you treat agents just like you would users.

[A]

Right.

[B]

So I think, you know, between that the whole idea that these autonomous agents have an entre ID is leading us down that path that they're going to be licensed in some way as users. On top of that, if AI is truly going to be reducing employees in companies, you got to look at it from Microsoft standpoint that their other subscriptions are going to drop in numbers. So to make up for that, it seems to make sense that agents will be licensed. They may not be licensed exclusively as a one cost fixed fee type thing. It may be a lower cost with, you know, some sort of usage charge on top of that, because that seems to be the direction that Microsoft is going in general. But yeah, this whole thing is going to be really interesting to see how it plays out.

[A]

It definitely is. Okay, so there also were some hardware announcements at Build besides the Android devices I mentioned at the beginning. The week before Build we had Computex and there, there were a lot of new Windows PCs announced that are based on Nvidia's ARM based RTX Spark. And at Build, Microsoft even announced a dev box that's going to be out in the fall that will be built on the same chip. So all these PCs, including a new very high end Surface that's coming, all look really nice. But I'm curious if you think enterprise developers are going to bite on Microsoft's argument that you need this kind of souped up high end hardware to run models locally. That seems to be their pitch on this, right?

[B]

Yeah, that Surface RTX looks so nice.

[A]

I can't wait to see the price tag. I'm sure it's.

[B]

Yeah, exactly right, Astronomical.

[A]

Yeah,

[B]

that's definitely going to be part of the equation, but I don't think it's going to be the developers that they have to sell. I think it's going to be the bosses. Looking at what they rolled out, I see two different business cases. Potentially the RTX Spark dev box that they announced was basically a high end box and targeting local model development. Not necessarily your typical developer. Everything indicated that it could be used for inference tuning training. So this is going to be a smaller target market, but one that benefits from using this kind of hardware versus the expense of using cloud GPUs. So that that could be a very viable business case for companies in that market. I think the broader developer market may not necessarily need the RTX Spark, but looking at the initial impact to changing GitHub Copilot's pricings, for example, devs would probably benefit from having something like the Ultra to run models locally and that way they're not incurring those additional AI costs. I mean, it's no longer an all you can eat buffet and it's driving up the cost of development again, something AI was supposed to reduce. And of course the other challenge in the mix is that enterprise hardware is usually on a two or three year cycle, so I don't know where that's going to go.

[A]

Yeah, interesting. Okay, this is a good part for me to take a break because I want to talk about something that we've had under wraps here at Directions for a while, our Atlas AI Assistant. We're now testing this with a select group of Directions on Microsoft members. We may be biased, but we think this is going to be the AI companion of your Microsoft licensing and technology dreams. We built it to make it simpler for you to find the enterprise focused Microsoft information you need based on the insights our team of experts have developed and curated over the past several years. Once we launch the Atlas AI Assistant in July, it will be available only to Directions on Microsoft members, not the general public. So make sure if you're not already a member, you check out the different ways you can join at directions on Microsoft.com there you can see the membership benefits that you could be making use of right now. Okay, back to David and our post Build ruminations. I have to talk about this. Even though it wasn't technically an announcement, it was a non announcement at build or at least a no show. But everybody was thinking Microsoft was going to announce the so called Copilot super app at Build. Satya Nadella, the CEO of Microsoft, acknowledged its existence during his keynote and said to expect it this summer. But that's all we got really. We didn't see a demo, we didn't get a lot of details. So let's talk about what we think think this super app might look like and what it potentially could help with. We're just blue skying it here. Let's go David, what do you say?

[B]

All right, well I have to say that unless we're talking like Star Trek's computer, I'm a little skeptical of the concept of a super app here. But if I were to blue sky this, I think computer is what I would target. It seemed to access personal data, understand intent and execute a variety of multi step tasks and it was generally well behaved. So what I'd be looking for is something along those lines, a proactive digital assistant. Something that was going to Help me stay informed and on track throughout the day, week and month, and able to assist with my tasks. I'd stopped short of a digital twin though, which might imply simulating my voice, tone and style because even computer got compromised once in a while.

[A]

Yeah, I don't know if they have big, that big an ambition for the super app. I don't know.

[B]

Maybe good to have goals, right?

[A]

I mean, what I have read and heard about this is the super app might be the way Microsoft tries to make it look like there's just one copilot. Right. So we have many, many different copilots in Microsoft. People got confused. They just seem to proliferate. They tried to tamp it down. They keep coming out with more of them. And I think Microsoft needs a way to say to customers, here's your co, here's your copilot workspace, or whatever they end up calling it, and here are all the different things you can do in that workspace. So maybe something like a tab metaphor on the top with like code and chatgpt. I mean, not, sorry, chat, not chatgpt, just chat copilot chat. And maybe something with these new always on agents as a tab. I think we're probably going in that direction. But my biggest question is, are they actually going to try to make this one experience for enterprises and consumers? Because to me, that seems like a really big bite to bite off.

[B]

Yeah. And I think you just tapped into why I'm skeptical of the super apps. I don't know if combining the consumer with the work is a viable goal at this point in time, at least not for V1. I think businesses would be really concerned about crossing the streams by having them both in there. And then the more realistic target, of course, is consolidating the copilot experience, which they've been trying to do with the Copilot app in general. I mean, you go in there and you've got lots of different things that you could do, but how do you make it useful? I think one of the challenges that they have is that Microsoft has been saying, we've been embedding this in places where you work to create the super app. You have to invert that whole premise and make the copilot app where you do work. What does that really mean? So you could potentially combine teams and Outlook and some of those types of tasks in one area along with your agents. But I look at it from the perspective of, all right, well, what if I'm working in a Word document or a PowerPoint? Are we talking about folding in the Office experience, which we have the Office agents in there, which, you know, so you could. But I then start to wonder about the whole prem. The whole idea of jack of all trades, master of none. Yeah. Are we really just going to get something that is mediocre across the board?

[A]

Right. More. More of a Franken app than a, than a super app. Yeah, yeah, yeah, I guess. We'll see. They say this summer. We'll see.

[B]

All right, stay tuned.

[A]

I know. Okay, now I want to turn our direction a little bit because after Build ended wrote a very interesting post for us for the Directions blog where you tied together how Microsoft's latest product and strategy directions that were amplified at build reflect its approach to AI. And you made it sound very simple and made sense. I was like, okay, I like your logic, but I feel like this is something many users may not yet have kind of pieced together. So let's talk about some of the things you raised in that post. So first off, this was, this is a bold contention, but you said ROI is the wrong lens when it comes to Microsoft's AI strategy. Now I think Microsoft would like that message because nobody can figure out how to calculate ROI on Copilot and the agents, including Microsoft. Right. So let's talk about what you meant when you said that.

[B]

Yeah, absolutely. I mean, I think people who, you know, not just Microsoft, but everybody's having trouble calculating.

[A]

True. That's fair.

[B]

ROI on AI. And it's going to get interesting because anthropic announced its IPO. OpenAI is expected to announce its IPO and they're going to have to figure out how to show revenue what the potential is. So I think ROI is in everybody's view as a metric for all of this. And because of the financial pressures everyone is experiencing, it's kind of like a short term expectation. It's like, okay, we've invested, we need to see what our returns are going to be on this investment. But Microsoft doesn't necessarily play that game, at least not all the time. And they are, you know, they're looking at the long haul here. So they're, they're willing, I think, to sacrifice the short term gains from Copilot and other AI initiatives in order to position themselves for the longer term shot at platform dominance. So it's a playbook they've used before. They've used it with Windows, they've used it with Office. Those seem to have worked out pretty well. And based on the Build conference, that looks like what they're angling to do.

[A]

Again, I don't disagree that they're always looking at the long term strategy view versus the short term. So another point you explored in your post is where lock in actually happens in Microsoft's AI stack these days. So what's changed here? Why isn't it what everybody thinks it is and what is it?

[B]

Yeah, I think the initial run at all this was all about the models. You had chatgpt come out, you had CLAUDE come out and Microsoft partnered with OpenAI. They were locked in with OpenAI so there was a lot of that focus on the models themselves. But in a sense models are almost a commodity at this point in time, which may sound a little strange, but they iterate very quickly. New models are coming out every, you know, much more frequently than they were when they first started happening. So I can switch out GPT for Claude or vice versa depending upon whichever model had particular capabilities or price point that I was looking for. But they're also basically generic. They don't have much value until they can actually leverage an organization's data. And a lot of that data is conveniently in M365. And that's where Microsoft's value proposition starts to come around. That's why they're introducing Microsoft IQ to provide that enterprise business context. Organizations are like, well if they're accessing our data, we want some control over them. Microsoft is stepping in to provide enterprise security and governance tools in the form of Entre ID and Agent365 and Purview. All of this is great, but none of it makes sense if the user can't access it or use it. It's soft is integrating it into everyday workflows. You've got teams and Office and Outlook and the soon to be fabled Copilot super app. There's a lot going on with that. But Microsoft is raising the bar and betting heavily on the platform itself rather than the underlying models.

[A]

Yeah, that makes sense. That's what they're good at, right? And that's where they're, that's their wheelhouse. So yeah, it makes sense. Okay, given what we know post build about Microsoft's evolving AI strategy, what's your advice? I'm going to put you on the spot. Should enterprises cobble together their own best of breed AI stacks using some Microsoft, some third party, maybe some of their own in house developed stuff or go all in or almost all in with Microsoft because if you do that it should be quote better together or are you going to pull an it depends as your answer to this. I'm curious.

[B]

I'll go with C. Mary Jo, final answer.

[A]

Okay,

[B]

so there's no one size fits all answer to this. We're still in a pretty early stage with a lot of things evolving, maturing and even disappearing. Expenses are rising. We're still struggling to show cause for AI in the first place, so enterprises risk tolerance is a big factor here. Microsoft is also changing the rules of the game this year. They're doing a lot to restructure agreements, they're increasing prices July 1st and at the same time they're signaling a shift to this cost plus usage model for AI and who knows what else. That's why understanding the potential for platform lock in is so important here. Companies really have to go into this eyes wide shut. Obviously, if you're a Microsoft shop, it may make sense technically to continue down that path, but your relationship has changed and costs are rising. So you want to be prepared to negotiate. If you're an organization that's concentrating on your line of business and you want to simplify the complexities of using AI, a platform approach like Microsoft's may make sense. And if you're in the business of AI, meaning you are actually building products and services with AI, then best of breed may make more sense. It depends.

[A]

All right, I'm going to let you get away with that because it makes sense to say it depends here.

[B]

Thanks.

[A]

Thank you so much David. That was great. We really appreciate the time and your insight. Thanks again.

[B]

My pleasure. Thanks for having me.

[A]

I would like to remind our listeners they can find lots more coverage of all things Microsoft related on directions on Microsoft.com thank you so much for listening. If you have questions, comments or any topics you would like to hear the directions analysts cover in one of these podcasts, please do not hesitate to contact me via X or a Blue Sky. Directions on Microsoft is on LinkedIn, so make sure you follow us there and give us a follow at DirectionsMSFT on X or directions on Microsoft on BlueSky for all of the latest Microsoft Enterprise product and licensing information. Thanks again.