Microsoft's AI Strategy and the new MAI Models

Microsoft is making a major strategic push to build more of its own AI capability — and business leaders should pay attention. In this episode of the Macro AI Podcast, Gary and Scott break down Microsoft’s evolving AI strategy under Mustafa Suleyman, including the company’s new MAI model family and how it fits into the broader Microsoft ecosystem. 

They explain the purpose of Microsoft’s new models: MAI-Thinking-1 for more complex reasoning, MAI-Code-1-Flash for developer workflows, MAI-Image-2.5 for image generation and editing, MAI-Transcribe-1.5 for turning audio into business data, and MAI-Voice-2 for voice, localization, accessibility, and customer experience. They also explain where Microsoft’s Phi family fits in as a smaller, efficient model layer for everyday AI tasks that do not require a large frontier model. 

The discussion focuses on why Microsoft’s strategy is about more than low-cost AI. It is about matching the right model to the right workflow, using Microsoft Foundry as a control plane for discovering, deploying, managing, and routing across models. Gary and Scott also cover where executives should look first — meetings and calls, software development, content creation, voice and localization, and complex reasoning — and why Microsoft’s existing footprint in Teams, Microsoft 365, GitHub, VS Code, Dynamics, Power Platform, Azure, and its partner ecosystem gives the company a major enterprise advantage. 

For CIOs, CTOs, CFOs, and business leaders, the key question is no longer, “What is the one best AI model?” The better question is, “What work are we trying to transform, and which model is the right fit?” 

https://microsoft.ai/models/

About your AI Guides

Gary Sloper

https://www.linkedin.com/in/gsloper/

Scott Bryan

https://www.linkedin.com/in/scottjbryan/

Macro AI Website: 

https://www.macroaipodcast.com/

Macro AI LinkedIn Page:  

https://www.linkedin.com/company/macro-ai-podcast/

Gary's Free AI Readiness Assessment:

https://macronetservices.com/events/the-comprehensive-guide-to-ai-readiness

Scott's Content & Blog

https://www.macronomics.ai/blog

Transcript

00:00
Welcome to the Macro AI Podcast,  where your expert guides Gary Sloper and Scott Bryan navigate the ever-evolving world of artificial intelligence.  Step into the future with us  as we uncover how AI is revolutionizing the global business landscape  from nimble startups to Fortune 500 giants.  Whether you're a seasoned executive,  an ambitious entrepreneur,

00:27
or simply eager to harness AI's potential,  we've got you covered.  Expect actionable insights,  conversations with industry trailblazers  and service providers,  and proven strategies to keep you ahead in a world being shaped rapidly by innovation.  Gary and Scott are here to decode the complexities of AI  and to bring forward ideas that can transform cutting-edge technology  into real-world business success.

00:57
So join us, let's explore, learn  and lead together.  Welcome back to the Macro AI podcast.  I'm Gary Sloper here with my cohost, Scott Bryan. Today we're gonna give business leaders a quick briefing on Microsoft's AI strategy. You've probably seen this in the news. uh Earnings came out uh absolutely crushing it. This is not just another episode about cheaper AI models. Microsoft is doing something bigger.

01:25
They're building a broader AI model portfolio, which models designed for reasoning,  coding,  transcription,  voice,  image generation.  Many of you use that and business workflows.  And they're doing it inside the Microsoft ecosystem.  So think of things like Copilot,  GitHub,  VS Code,  Microsoft 365,  Azure. Many of you work in that cloud environment and also Microsoft Foundry.

01:54
person at the center of this story is Mustafa Suleyman, the CEO of Microsoft AI. And you probably have seen some interviews with him. ah He co-founded DeepMind and he later co-founded Inflection AI. And now he's leading Microsoft's push to build more of its own AI capability directly. So today we're gonna explain what Microsoft is trying to do, what the new oh MAI models are.

02:22
where the five family fits in,  why the strategy matters,  and where executives like yourself should start paying attention in this  whole Microsoft shift. So Scott, think the big point is that Microsoft does not want AI to be a single giant model used for everything. They're building toward a  portfolio where different models do different jobs. Yeah,  exactly. uh

02:48
I think they were smart to sit a little bit on the sideline and focus on where it fits into their business.  And I think executives need to understand the strategic shift here. Because for years, Microsoft's AI story was closely tied to open AI, while Microsoft executives were thinking about exactly where to deploy capital and technically what to do here with their strategy.  And that open AI strategy is still extremely important, but Microsoft is clearly moving more ah

03:18
direct control over its own AI stack. ah So what does that mean? That means more control over model performance, more control over cost, control over product integration and how it fits exactly into their ecosystem  and more control over how AI  shows up inside that ecosystem with Microsoft products. Yeah, that's good points. And I think that's why  Mustafa Suleyman matters. I mean, he gives the strategy a lot of credibility.

03:48
This is not just Microsoft casually experimenting with a few models. Mustafa was one of the co-founders of DeepMind, like I mentioned a little while ago, and he later founded Inflection AI, and now he's leading  Microsoft's AI strategy here. So  when Microsoft releases its own MAI models, business leaders should see that as a strategic signal. Microsoft is saying, we're not just going to rely on one partner or one general purpose model, we're going to

04:17
build more of our own AI capability and integrate it directly into the Microsoft ecosystem. Yeah, exactly. And Microsoft has described this building of their own models as building a hill climbing machine, which is kind of an interesting phrase. I think the idea is that AI is not a  one-time product release. It's a system that keeps improving.  The models improve and each specific product improves. And then the whole feedback loops. uh

04:47
feedback loops improve and the routing gets smarter and the use cases become more specialized. And that really matters for each enterprise that's  deploying inside their ecosystem. Yeah, good point. mean, that is important because enterprise AI is becoming really an operating layer. It's not just a chat bot sitting on the side. And we've talked about that in other episodes.  Microsoft wants AI inside the actual work, whether that's  through meetings, documents, coding.

05:15
your sales team, service,  creative production, data workflows, uh and just overall business applications.  And the key point is that Microsoft is not treating all of those as the same problem. That is why  the new MAI model family matters.  So maybe we dive a little bit more into  the models at a high level. I Microsoft recently announced a family of in-house MAI models. These are...

05:43
covering things  such as reasoning,  code, image, transcription, and voice. Think about what they do with just Microsoft Teams today. So  maybe  if we start with the reasoning model,  it is really one of the, that  people will probably associate with harder business problems within their own organization. Yeah, I can take that. So the reasoning model is uh MAI thinking one.

06:13
And I think the simple way to explain it is that  Microsoft's AI model,  it would be their model for harder work. So not every routine task or simple summary, but more for the difficult problems that require multi-step reasoning. So a little bit more similar to the big frontier models. So think about complex analysis, planning,  really tough software engineering tasks,  math heavy work or policy, ah or even

06:42
decision support workflows. So where the model needs to work through a problem  step by step and work through it carefully. So the strategic point that Microsoft here is saying is what they're trying to say is that ah not that we built the biggest model,  but we're building strong reasoning capability in a more efficient architecture designed specifically for our enterprise users.  that matters because the future isn't just

07:12
lots of raw intelligence. It's intelligence at the right cost inside the right workflow for your business. Yeah, that's a really important distinction. So MAI thinking one is the model you escalate, you know, to when the work is more complex as you just alluded to, right? It's not the model you necessarily use to summarize every meeting or classify every support ticket that you might have used in other models. The next model in

07:40
is  the next model that they have is MAI code one flash. And this one is a major Microsoft ecosystem story. This is the coding model built for developer workflows. So Microsoft has a huge advantage here because it obviously owns GitHub. GitHub Copilot is already part of how many developers work and VS code is one of the most widely used developer environments in the world.  So.

08:05
So MAI code one flash is not just another coding model floating around in the market is designed to fit into GitHub, co-pilot, VS code and the Microsoft developer stack. So if you're in that environment, this is, this is a huge announcement. Yeah, it's right there and it's top of line and, and, and that's where business value comes in. So companies are, they're already using AI to generate tests, you know, work on code, refactor code, debug problems. And then now, now they're starting to work on  more complex.

08:34
agentic software development. And if Microsoft can make that faster, cheaper, and more tightly integrated into the tools that developers are already using, that's a really solid productivity story. uh And I think it also matters because software development is one of the places where AI usage can really become constant. So developers might use AI  all day long. So efficiency matters.  So I know I've run up some costs doing this.

09:04
And I think in Microsoft's own benchmark results, uh MAI code one flash beats Claude Haiku four or five across the coding tests that they publish. So, you know, that's just a demonstration that it's  solidly capable offering. Yeah. The benchmarking results were interesting and a positive start for Microsoft. So  if we move to the creative side.

09:29
Microsoft also has MAI image 2.5, including a flash variant. So this is for, you know, text to image generation and image editing. The business use cases  can be associated with marketing concepts, sales enablement. uh You know, if you're doing a product mock-up, presentation visuals uh could also be, you know, for LMS training content  and overall internal communications, just that kind of that branding. Yeah, all branded.

09:57
Yeah. And, and so, you know,  the, again,  the Microsoft ecosystem matters,  especially for a lot of enterprises.  So if image generation becomes more connected to things like PowerPoint, which so many people use, or, you know, Microsoft 365 and, and just corporate content workflows, it becomes more than non-novelty. It becomes part of the way business content gets produced, which can be a real game changer for the enterprise. Yeah.  100%, especially like you said,  when it's all

10:27
branded and built into your environment.  So I think for executives, point is not, know, AI can make pictures.  The point is that content production is  generally it's expensive and slow and marketing teams,  sales teams, training teams, product teams, they constantly need visual material.  And  like we said, that has all the latest branding. So if AI can  speed up early drafts, concepts,

10:56
localization and all the visuals that can really improve the whole content pipeline across all those  eight or so use cases we just talked about. Yeah. And I've even seen  smaller organizations think of like pizza shops and others that are just starting to, you can tell it's AI image generation, but it's  increased just the content for them. So imagine what that could be at the hands of an enterprise just through Microsoft. It's amazing. uh

11:24
So the next model is MAI Transcribe 1.5. And I think this may be the easiest one for executives to understand. So every business has audio,  whether it's... Yeah,  lots of audio. Lots of audio. So think of Teams meetings,  you're on sales calls,  so you have your SDRs and BDRs.  You have a lot of CX needs,  customer support calls,  contact center recording, and even just training sessions or field service notes, et cetera.

11:53
your organization does for your business.  Microsoft says MAI Transcribe 1.5 is built for accurate multilingual transcription across 43 languages out of the box.  And once audio is transcribed quickly and accurately, it becomes usable business data.  So think of that, just the power that you have at your hands.  And all of that audio transcribed into business data.  Yeah,  it's super powerful.

12:20
latency across your private network can be designed to optimize the performance here. So now you're taking something that, you know, again, kind of almost off the shelf and you're immediately becoming, you know, a multilingual organization out of the box. Right. Yeah. And so I think just kind of breaking that down a little bit so that the transcript is really just the beginning. Then after that AI can summarize all those calls, extract all the action items, update CRM. uh

12:49
It can talk about uh detect and analyze customer sentiment across any number of things. They can identify product feedback and summarize that to each individual product team. Pretty amazing. ah It can listen to and then flag compliance issues  and then all the stuff that people are familiar with, creating follow-up emails or whatever. So definitely a lot that can happen there.  And I think this is where these models start to turn ordinary business conversations into

13:18
structured business intelligence,  and then obviously that just keeps compounding  from here on out. It does, it does. And I would say, I almost feel like Steve Jobs. And finally, there's one more thing. ah MAI Voice 2. So this is Microsoft's speech generation model. The opportunity here is voice-based interfaces. So think of, know, multilingual content, training, accessibility, customer support, which I mentioned before, and localization.

13:48
So  what does this all mean? So for global companies, that can be powerful. You can imagine training content and customer messages and internal communications, whatever that might be that you need, all being produced across languages much more efficiently than you've ever been able to do before. Yeah. So not taking voice and transcribing, but actually generating voice. yeah,  good stuff. So if we kind of, if we step back again, the  MAI family really maps to

14:16
real categories  of enterprise work.  was pretty,  looking back on it, I think it was a smart strategy. Everybody's wondering where Microsoft was as  these frontier models are being built out. But now they've got these really capable models for reasoning, code, image, transcription, voice. And I think that is a strategy.  Microsoft is building a portfolio of models for the way that businesses actually operate. And obviously they...

14:44
They know this, they have hundreds and hundreds of thousands of business customers. Yeah, and to that point, I think that is  the business executive takeaway. This is not just one model attempting to do everything. It's a model family designed around the major way, you know, businesses use AI today. Right. Yeah, and we, long time ago, we talked about one of the early models, uh FI, and we'll just, talk a little bit about... uh

15:11
I think we should take a minute and just explain the FI family because it's different from the MAI models. Yeah. And it's important. FI is Microsoft's family of small language models. So instead of thinking about FI as another giant frontier model, uh think of it as an efficient model layer. These models are designed for jobs where you want useful AI capability, but you don't want to necessarily need the biggest model in the world. That could be things like summarizing short documents.

15:40
running internal assistance, helping with lightweight reasoning or powering AI features closer to the edge for your business. Yeah. And I think the reason that by matters is that a lot of business AI work is really repetitive and high volume.  Um, and it's fairly easy to customize these  spy small language models and they've been working on them for a while. Uh, these, one of the older ones, like we mentioned. So if you're, if you're processing thousands of documents, calls, tickets, whatever,

16:10
and you don't want every step going to the most expensive model. uh PHY is relatively easy to deploy and it gives uh Microsoft a lower cost layer for work that needs to be fast, efficient, and good enough for the task. Right. So I guess the way to think about Microsoft strategy is really in layers.  PHY and other smaller models for efficient everyday tasks, ah MAI specialized models.

16:38
for things we talked about transcription,  voice,  image,  coding,  reasoning.  Foundry is the platform to deploy,  compare,  manage and write across models.  And then, you know, many of you are on Azure.  Azure as the infrastructure layer underneath it all.  Yeah. Yeah. And on Azure,  when you see the, if you dig into today's announcement about their earnings,  a lot of growth there because models are being deployed there.

17:08
And I think just overall, point is that, know, FI is not replacing the MAI models. It gives Microsoft a smaller, more efficient layer that's still going to be in the portfolio and they're going to continue to build on that. So, so businesses can really specifically match the model and the model type to the work.  And like you said, instead of using one big expensive frontier model for everything. And I think that brings us.

17:35
to the broader strategy, right? The strategy is workload matching. The important idea is that Microsoft is not telling companies to use one model for everything. They are giving businesses more ways to match  the model to the workload. So for transcribing audio using a  transcription model, for developing code using a coding model. If you're generating a virtual uh asset, then you have an image model.  We talked about that a little while ago. If you're doing routine,

18:05
uh, classification or summarization using a smaller, lower cost model. If you're performing hard reasoning using a stronger reasoning model, it sounds obvious, but a lot of the early AI deployments were not built this way. And I think that's where this is a very interesting shift. Yeah. And yeah, definitely. I think a lot of companies took one powerful general purpose model and started sending everything to it. It was just, you know, phase phase one.

18:34
And that works for  the demos and the demo phase, but in production and as your IT teams get smarter and you're looking at things at enterprise scale, using just those expensive models is obviously expensive and inefficient.  that the future is  obviously model specialization, like we've been talking about, and Microsoft is  building toward that future for the enterprise architecture. Right.  Right. So.

19:04
For executives using the Microsoft ecosystem, the question should not be, what is the one best model? This is a little bit different.  So the question really should be, what are our AI workloads and which Microsoft model or service is best fit for each one? And for the CFOs, it's time to include AI as part of your FinOps practice.  All of this represents a shift from AI experimentation to AI operations.

19:33
We kind of talked about this a little bit in our last episode with E-gain because we are seeing this go into full production.  exactly. And to your point about AI pin-ops,  Microsoft's model strategy gives companies more ways to control AI economics without just saying no to AI. Right.  So you're not cutting AI usage, you're making AI more operationally mature. And I think that's the reason why

20:02
for the Macro AI podcast, we're starting to get a lot more CFO listeners because they  need to have a better understanding of  really what we're talking about is how all these things fit together. Well, and they're getting asked by the board, right? Often, you and so you make a really good point around the AI maturity. That's where the board wants to understand,  are we still playing around with tools and  maybe thinking about models and other things, or are we actually putting this into

20:30
you know, an operational fixture for the business. Yeah, exactly. Um, all right, let's just shift a little bit over to kind of the control plane of how this all comes together. And that's, that's be Microsoft Foundry.  Um, cause this is where the model strategy really becomes usable for businesses.  Um, so Foundry is the platform layer where companies can, can go in and discover models. can deploy models, compare them,  um, build AI applications.

21:00
and manage access to different model options.  And  then  on top of that, one of the most important concepts is model routing. And I think we'll have an episode specifically around model routing coming up. Yeah, that's a good idea because  model routing is critical. Instead of hard coding every application to just one single model, you create a routing layer. The system can decide which model should handle a request based on cost.

21:30
quality, complexity, ah and also the nature of the work. If the workflow is simple and high volume, you may route to a lower cost model. If it's complex or high risk, you may route to a stronger model. And you've probably experienced this with some of the public LLMs. You start burning through tokens and you're like, maybe I should have gone down a little bit. ah If it's  legal, medical  related or customer facing, you may route it through a more controlled workflow with compliance.

22:00
Yep. This, is a very interesting topic. Yeah. And then specifically, uh, Microsoft Foundry supports, uh, routing options.  So a little bit more technical, you can routing options would include,  uh, balanced,  uh, cost and quality. And that's a kind of a high level, uh, executive friendly concept, I guess you could say. So balanced says,  uh, give me a good mix of both quality and cost.  Um, so the, the cost.

22:30
routing options says for this particular workload, prioritize efficiency. And then the last one is that I mentioned is quality. And that one would be, you know, for this workload, I care more about the best answer. Yeah, that's a point. And that is how AI becomes manageable because agents can multiply model calls. So think of it this way, a user asks one question, but behind the scenes, the system may search  a lot of different areas. It'll search documents, call tools.

23:00
summarize information,  it'll rewrite the output,  and ultimately format the final response. If every step goes through the most expensive model, the economics can get ugly fast,  and your CFO will start asking questions about rising costs.  if a router sends simpler steps to cheaper models and reserves stronger models for harder steps,  companies get a much better operating model.  And you alluded to

23:29
to the whole FinOps strategy, Scott, that that's,  you know, part of, you know, managing your environment here. It's not just about saving costs. It's about the efficiency in this case for your model. Right. Yeah. And I think this is where AI starts to look like a cloud infrastructure. So you don't use the, you know, the most expensive compute or storage for every workload. You match the infrastructure to the job and AI is really just starting to go through the same transition as the

23:59
uh architectures start to settle out. Right. So  where executives should see the opportunity and you and I came up with a list before this recording. So maybe we kind of kind of go through that because for our executives listening that want to know where this Microsoft strategy could matter, we would start with probably five areas to start with.

24:22
ah So the first one that I had on the list was meetings and calls. So MAI Transcribe 1.5 is highly relevant here. So think about your team's meetings or customer calls or support calls,  contact center recordings. The opportunity is not just transcription, which we talked about. It is turning all of that spoken information into searchable, summarized, actionable business data. Very valuable business data that you can consume and do a lot of things with. Yep.

24:52
Yep, definitely.  And then the second one would be software development. That's obviously a hot topic. So MII  code one flash fits directly into Microsoft's GitHub and VS code advantage. And companies should expect Microsoft to really keep pushing AI deeper into the developer workflow.  And that matters because software development is really becoming one of the most AI intensive areas of the business. And you can bet that

25:20
Microsoft is positioning itself to, to compete with Anthropics Cloud on capability and importantly on, on cost. So I think they will end up with a solid advantage here. If not, if not on par, maybe an advantage. Yeah, there'll definitely be a head to head battle for sure.  Um, so third, one that we came up with is content and creative production. So MAI image 2.5 matters for marketing matters for sales enablement training.

25:51
internal communications and presentation heavy work, which we kind of alluded to earlier. So it's not just about generating images. It's about speeding up the content cycle, custom  and up to brand standards that you have and expect within your marketing team as an organization. Yeah. Huge productivity. uh So fourth, voice and localization. So MAI Voice 2 can support training, accessibility, multi-language,  multilingual communication.

26:20
uh Any kind of voice-based interface and for  global companies, localization is a huge opportunity when you've got multiple languages happening across your entire enterprise. This is a no-brainer and it's not that difficult to deploy. Yeah, good point. And so the fifth would be complex reasoning, which I'm sure you probably expected to be in the list. ah MAI Thinking 1 is the model to watch for harder analysis, planning, coding.

26:49
policy interpretation and,  really decision support workflows.  Yeah. And you know what? I just want to add,  was the fifth post at one executive filter across all five.  So look at where Microsoft already owns the workflow in your company.  So if your company already uses, you know, teams, outlook,  SharePoint,  365,  GitHub,  lot of are using more complex things like dynamics,  power platform.

27:18
So  Microsoft has a natural path to bring these models into your environment. So that doesn't mean every company should only use Microsoft models, but it does mean that Microsoft may be the most practical path for a lot of enterprise AI adoption. And you can bet that a lot of the uh Microsoft  value-added partners out there are going to be competing for business and you're going to have to assess them for their skills for what you're looking to do. point. And a model is valuable, but

27:48
distribution matters. does, you know, integration as well and security and governance and most of all identity matters. So Microsoft's advantage  is that they already sit inside the enterprise stack and you're using this today. So just keep that in mind as you're doing some of this analysis.  Yeah. Let's, just kind of jump into the kind of the hosting and deployment options. uh So let's touch briefly on deployment. ah So for executives listening, you don't need to become

28:18
infrastructure engineers, but you should understand that you have a lot of options. So the easiest path is to use models through Microsoft Boundary and Azure, and that keeps the operational burden lower because  Microsoft handles much of the platform infrastructure. And for many companies, that's an easy place to start, but there are lots of other options. Yeah. And for companies that need even more control, self-hosting or co-location would be part of the conversation. I know, Scott, you and I both- Yeah, we're-

28:47
We're starting to that quite a bit and the CFOs are starting to ask,  can we save money by going  colo? Colo, like on a hybrid model. So you have your on-premise, we'll call it on-premise with a third party data center and your Azure environment. uh And they're high quality colocation data centers in virtually every major business market in the world. So companies that want say a private AI infrastructure close to users, cloud regions or corporate networks usually have multiple options.

29:16
And so I do a lot of this with companies today that use network as a service to connect to the likes of Microsoft and as well as Oracle cloud, for example. Um, and I'm seeing that huge shift where we thought, you know, the co-location requirements were continuing to go down and now they're shifting,  you know, increasingly up. Yeah. And I'll just add to that, that there were starting to figure out that there are quite a few more network as a service options that are.

29:46
that are valid as well, making networking a lot easier than it has been in the legacy telecom environment. So that's really interesting to see too.  And  then again, with the explosion of capable open weight, open source models, I expect a lot of OPEX to shift to that colo model and then networking to once again become a topic on the CIO's radar because there are even other ways to cut costs and build your own secure environment.

30:16
Yeah. And, and, and a lot of the networks that exist today were built for 24, 36 plus months ago when AI adoption,  even if you're using, you know, traditional frontier models, uh, they didn't exist. didn't need the bandwidth.  just, just something else to keep in mind. Yeah. think a lot of those, like you said,  36 months ago, a lot of the wide area network stuff is just simply renewed. But now is probably, now is probably a good time to take a, take another look at it. Yeah.

30:44
Yeah, spend a lot of time helping companies benchmark their pricing and strategy for sure. And you're probably thinking, so as a business leader, what should you do next? And if we were to bring this all home, if  you are an executive listening to this, I would do four things.  First, ask your technology team for a Microsoft specific AI briefing, not just a generic AI update. Ask specifically about MAI models, buy models, Microsoft Foundry.

31:10
and model routing. think those are some key areas to hone in on. And it doesn't necessarily mean the team's doing a bad thing if they're not focused on it yet, but  now you're more aware to probe and help coach the team to go a little bit deeper if they're not in them.  Exactly. Yeah, the strategy is becoming more clear. uh So  second and a little deeper there, I identify where Microsoft already  owns the important workflows across the business. And we talked about that. So Teams, Outlook, SharePoint.

31:39
Dynamics power platform, where Azure sits, things like that. Yeah, yeah, good point. And third, map the model family to real use cases.  So, transcriptions for calls and meetings, uh code models for developers, image models for marketing and training, voice models  for localization and accessibility, ah five models for efficient everyday AI tasks, reasoning models for harder analysis and decision to support all the

32:08
the core areas that we talked about at the beginning of this episode. Yeah. Yeah. And think there's a fourth one after that and ask whether your AI architecture can route work to the right model or whether everything is being sent to uh expensive general purpose models.  So that fourth point is critical because Microsoft strategy only helps if companies build in a way that they can take advantage of it. uh

32:35
So if your AI application is hardwired to one model, you might miss the benefits of, you know, the overall model portfolio. uh But if your architecture can, can, can test, can route, it can swap models. You can take advantage of new Microsoft models and outside models as they improve.  And don't look at these announcements as isolated model releases. This is a signal about where Microsoft is going.

33:02
more models, more specialization, more integration, more routing, ah more enterprise controls. So  I think in summary, Microsoft is building more of its own AI capability under  Mustafa. That is strategically important. And  the new MA model family is built around real categories of work, reasoning, coding, image generation, transcription and voice. Also, ah

33:31
Keep in mind, FI is Microsoft's efficient small model layer for everyday AI tasks that essentially do not need the biggest model in the world. So just keep that in mind when you're thinking about that FEN Ops approach. Yeah. And just one other takeaway uh would be that Microsoft Foundry is becoming the control plane  for discovering, deploying, managing, running across the models. And I guess there is one more point. uh

33:58
Microsoft has a major distribution advantage over the other major AI labs because they already have thousands of enterprise customers and certified partners. So if they keep rolling out great model options that are very cost effective, they're obviously positioning them well and something you should be taking a look at if you're in the Microsoft ecosystem. So the question for business leaders is not what is  the one best AI model? The better question is.

34:26
What work are we trying to transform and which model is fit in that scenario? That's really a strategic conversation. So this was a great discussion, Scott. I really appreciate it. Yeah. Yeah. Thanks, Gary. That was good. Thanks everyone for listening to the Macro AI podcast. If this episode was useful, share it with someone who wants a clear view of where Microsoft is going with AI.  Until next time, we'll see you soon.