Latest / Ctrl AI Profit / Ep. 127 | Microsoft Just Divorced OpenAI — And It Changes Every Microsoft Shop
Transcript
- 0:20SPEAKER_01: Microsoft just told OpenAI, thanks, but we've got this.
- 0:24SPEAKER_01: They're ripping GPT 4 out of GitHub Copilot and replacing it with their own AI model.
- 0:31SPEAKER_00: Project Polaris, Microsoft's in-house coding model, takes over as the default for every copilot subscriber starting in August.
- 0:39SPEAKER_00: 1.8 million developers are about to have the AI writing their codes swapped out from under them.
- 0:45SPEAKER_00: Let me put this in perspective.
- 0:47SPEAKER_01: This isn't some side feature.
- 0:48SPEAKER_01: GitHub Copilot is the most widely used AI coding tool on the planet.
- 0:53SPEAKER_01: It's in the editor that most developers open every morning.
- 0:56SPEAKER_01: Changing the AI brain inside it is like Ford swapping out every engine in every car on the road and saying, trust us, it's better.
- 1:06SPEAKER_01: This is the biggest product divorce in tech.
- 1:09SPEAKER_01: Microsoft and OpenAI have been the power couple of AI for years.
- 1:14SPEAKER_01: Copilot was their baby.
- 1:16SPEAKER_01: And now Microsoft is saying, I can do better on my own.
- 1:21SPEAKER_00: To be fair, they're not cutting ties completely.
- 1:24SPEAKER_00: OpenAI models will still be available in Copilot as an option.
- 1:28SPEAKER_00: But the default, the model that runs when you just hit tab and accept the suggestion, that's Microsoft's model now.
- 1:35SPEAKER_00: And defaults matter.
- 1:36SPEAKER_00: Most people never change them.
- 1:38SPEAKER_01: Here's why this matters for every small business owner listening.
- 1:41SPEAKER_01: If you use Microsoft tools, and let's be honest, most businesses do, the AI inside those tools is changing.
- 1:49SPEAKER_01: Word, Excel, Teams, Outlook, they're all getting copilot.
- 1:53SPEAKER_01: And the AI powering all of it is shifting from open AI's models to Microsoft's own.
- 2:00SPEAKER_00: And that shift affects you in three ways price, performance, and privacy.
- 2:06SPEAKER_00: Let me break each one down.
- 2:08SPEAKER_01: Before you do, I want to be clear about who this affects.
- 2:11SPEAKER_01: You might be thinking, I don't use GitHub Copilot, so this doesn't matter to me.
- 2:16SPEAKER_01: Wrong.
- 2:16SPEAKER_01: Copilot is just the beginning.
- 2:18SPEAKER_01: Microsoft is building their own models for everything.
- 2:21SPEAKER_01: Word, Excel, PowerPoint, Teams, Outlook, they're all getting copilot features.
- 2:27SPEAKER_01: And all of those features are moving toward Microsoft's own models.
- 2:31SPEAKER_01: If you use Microsoft 365 in your business, this is your story.
- 2:37SPEAKER_01: Price first.
- 2:38SPEAKER_01: Why would Microsoft spend billions building their own model instead of just paying OpenAI?
- 2:44SPEAKER_00: Margin.
- 2:45SPEAKER_00: When Microsoft uses GPT, for in Copilot, they pay OpenAI for every single API call.
- 2:52SPEAKER_00: Every time Copilot suggests a line of code, OpenAI gets a cut.
- 2:57SPEAKER_00: When you have 1.8 million subscribers making thousands of suggestions a day, that's a massive check Microsoft is writing to OpenAI every month.
- 3:05SPEAKER_00: With Polaris running on Microsoft's own Maya chips in Azure, that cost drops dramatically.
- 3:10SPEAKER_00: Microsoft keeps the margin.
- 3:12SPEAKER_01: And when Microsoft keeps the margin, they can do two things.
- 3:16SPEAKER_01: They can lower the price for customers, or they can keep the price the same and make more money.
- 3:21SPEAKER_01: Guess which one is more likely?
- 3:24SPEAKER_00: History says they keep the price in pocket the margin, but there's a twist.
- 3:28SPEAKER_00: Lower AI costs mean Microsoft can afford to put copilot into more products at lower price points.
- 3:34SPEAKER_00: Copilot for small business at$10 a month instead of$30?
- 3:38SPEAKER_00: That becomes economically viable when you're not paying OpenAIs per token pricing.
- 3:43SPEAKER_01: So the small business owner who's been priced out of Copilot might suddenly find it affordable.
- 3:48SPEAKER_01: That's the upside.
- 3:50SPEAKER_00: Now performance.
- 3:52SPEAKER_00: Microsoft says Polaris outperforms GPT for turbo on coding benchmarks.
- 3:58SPEAKER_00: They're using a mixture of experts architecture.
- 4:00SPEAKER_00: Basically, instead of one massive brain, it's a team of smaller specialized models that activate depending on the task.
- 4:07SPEAKER_00: That makes it faster and cheaper to run while matching or beating the quality.
- 4:12SPEAKER_01: Think of it like this GPT, for is a general surgeon, Polaris is a team of specialists.
- 4:18SPEAKER_01: For most routine cases, the specialists are faster and cheaper.
- 4:22SPEAKER_01: But for the weird edge cases, the stuff nobody's seen before, the generalist sometimes has better instincts because they've seen more variety.
- 4:31SPEAKER_01: But here's the question nobody's asking.
- 4:33SPEAKER_01: Outperforms on benchmarks, sure, but benchmarks aren't your Tuesday afternoon.
- 4:39SPEAKER_01: Your code base isn't a benchmark.
- 4:41SPEAKER_01: Your legacy system with the weird API and the custom framework isn't in the test suite.
- 4:47SPEAKER_00: That's the real risk.
- 4:48SPEAKER_00: When Copilot switches models in August, the suggestions that were working great for your specific code base might change.
- 4:55SPEAKER_00: The patterns the AI learned from GPT for might not translate exactly to Polaris.
- 5:01SPEAKER_00: It's like switching from a mechanic who's been working on your car for two years to someone new.
- 5:06SPEAKER_00: They might be better overall, but they don't know your car's quirks yet.
- 5:10SPEAKER_01: So what should businesses do about this?
- 5:12SPEAKER_01: Just be aware, when August hits and Copilot feels a little different, suggestions that used to be perfect are now slightly off, or the AI seems faster, but maybe less nuanced.
- 5:23SPEAKER_01: That's not your imagination.
- 5:25SPEAKER_01: That's a model change.
- 5:27SPEAKER_00: Microsoft is giving subscribers a three-month window to opt back into GPT.
- 5:33SPEAKER_00: If your team relies heavily on Copilot and the new model feels like a downgrade for your specific workflow, flip that switch.
- 5:41SPEAKER_00: Don't just accept the change because it's the default.
- 5:52SPEAKER_00: Especially businesses and regulated industries, healthcare, finance, legal.
- 5:57SPEAKER_00: If you've been holding off on AI tools because you're worried about where your data goes, this changes the picture.
- 6:03SPEAKER_00: When Copilot runs on GPT, four, your code goes to OpenAI's infrastructure.
- 6:09SPEAKER_00: When it runs on Polaris, your code stays in Microsoft's Azure, same company that hosts your email, your documents, your Team's calls.
- 6:17SPEAKER_00: Your data goes from a two-party system, you and Microsoft plus OpenAI, to a one-party system.
- 6:26SPEAKER_00: Just you and Microsoft.
- 6:28SPEAKER_01: For a lot of businesses, that's actually an improvement.
- 6:31SPEAKER_01: Fewer companies handling your data means fewer places it can leak.
- 6:35SPEAKER_01: And Microsoft's enterprise compliance certifications, HIPAA, SOC2, all of that, already cover Azure.
- 6:43SPEAKER_01: You don't have to separately verify OpenAI's compliance.
- 6:47SPEAKER_00: But there's a flip side.
- 6:49SPEAKER_00: Consolidation means concentration.
- 6:51SPEAKER_00: When one company handles your email, your documents, your AI, your code, your infrastructure, that's a lot of eggs in one basket.
- 7:00SPEAKER_00: If Microsoft has a breach or an outage or a policy change, it hits everything at once.
- 7:07SPEAKER_01: I've seen businesses get burned by vendor concentration.
- 7:11SPEAKER_01: The company that ran their email, their calendar, their CRM, their billing, all one platform.
- 7:17SPEAKER_01: When that platform went down for two days, the business went down for two days.
- 7:22SPEAKER_01: Nothing, no workaround, no fallback.
- 7:25SPEAKER_00: So the privacy improvement is real, but it comes with a concentration risk.
- 7:30SPEAKER_00: Small businesses should have at least one critical system, even if it's just backups, running on a different platform.
- 7:36SPEAKER_00: Not because Microsoft is untrustworthy, but because single points of failure are always dangerous.
- 7:42SPEAKER_01: Let me zoom out for a second.
- 7:44SPEAKER_01: What does this divorce tell us about where the AI industry is going?
- 7:49SPEAKER_00: It tells you that the era of AI company sharing is ending.
- 7:52SPEAKER_00: In 2023 and 2024, the story was partnerships.
- 7:57SPEAKER_00: Microsoft plus OpenAI, Google plus Anthropic, everyone cooperating.
- 8:03SPEAKER_00: Now every major player is building their own stack.
- 8:06SPEAKER_00: Microsoft has Polaris.
- 8:08SPEAKER_00: Google has Gemini.
- 8:10SPEAKER_00: Amazon is building their own models.
- 8:12SPEAKER_00: Meta just launched Superintelligence Labs and cut 600 jobs to restructure their entire AI division around it.
- 8:20SPEAKER_00: Even Apple is integrating Gemini directly into Siri instead of building a chatbot layer.
- 8:25SPEAKER_01: The partnerships were never about cooperation.
- 8:28SPEAKER_01: They were about speed.
- 8:30SPEAKER_01: These companies needed AI features fast and didn't have time to build from scratch.
- 8:35SPEAKER_01: Now they do.
- 8:36SPEAKER_01: So they're building their own.
- 8:38SPEAKER_00: And for small businesses, this fragmentation means something specific.
- 8:42SPEAKER_00: Your AI tools are going to multiply, and they're not all going to talk to each other.
- 8:47SPEAKER_00: The AI in Microsoft Word works differently from the AI and Google Docs, which works differently from the AI in Slack.
- 8:54SPEAKER_00: You're not choosing one AI, you're choosing an ecosystem.
- 8:58SPEAKER_01: Think about it like the old Mac versus PC divide, but way more consequential.
- 9:04SPEAKER_01: In the 90s, you picked a computer and most of your software came from that ecosystem.
- 9:10SPEAKER_01: Now you pick a cloud platform and your AI comes with it.
- 9:13SPEAKER_01: The stakes are higher because AI isn't just a tool, it's an employee.
- 9:17SPEAKER_01: An employee that learns your business, gets better over time, and doesn't transfer to a competitor's platform.
- 9:24SPEAKER_01: That's the decision every business owner needs to make this year.
- 9:27SPEAKER_01: What ecosystem am I in?
- 9:29SPEAKER_01: Because the AI that comes with your ecosystem is the AI you're going to get by default.
- 9:35SPEAKER_01: If you're a Microsoft Shop Teams, Outlook, SharePoint, Azure, you're getting Polaris AI.
- 9:41SPEAKER_01: If you're a Google Shop workspace, cloud, Android, you're getting Gemini AI.
- 9:46SPEAKER_01: If you're split across both, you're getting two AIs that don't coordinate.
- 9:51SPEAKER_00: And that's the hidden cost nobody talks about.
- 9:54SPEAKER_00: Running two AI ecosystems means training your team on two sets of behaviors, two sets of prompts, two sets of limitations.
- 10:02SPEAKER_00: It's not twice the work.
- 10:03SPEAKER_00: It's more than twice because you also have to manage the gaps between them.
- 10:07SPEAKER_01: My recommendation, pick a side.
- 10:09SPEAKER_01: If you're already mostly Microsoft, lean in.
- 10:12SPEAKER_01: If you're already mostly Google, lean in.
- 10:14SPEAKER_01: Don't try to be agnostic.
- 10:16SPEAKER_01: The businesses that commit to one ecosystem will get the most out of the AI that comes with it.
- 10:21SPEAKER_01: The ones that try to stay neutral will get the worst of both worlds.
- 10:24SPEAKER_01: Fragmented AI, inconsistent workflows, and a team that's always context switching.
- 10:30SPEAKER_00: Think about what happens when your AI assistant in Outlook knows your writing style, but the AI in Google Docs doesn't.
- 10:37SPEAKER_00: You write an email in Outlook and the AI finishes your sentences perfectly.
- 10:41SPEAKER_00: You switch to a Google Doc and you're starting from scratch with an AI that doesn't know you.
- 10:46SPEAKER_00: That friction adds up.
- 10:48SPEAKER_00: Every day, every document, every task.
- 10:52SPEAKER_00: And commit early.
- 10:53SPEAKER_00: The AI features in these ecosystems are compounding.
- 10:56SPEAKER_00: The more you use Microsoft's AI, the better it learns your organization's patterns.
- 11:01SPEAKER_00: The more you use Google's, the better it understands your documents and workflows.
- 11:05SPEAKER_00: That learning doesn't transfer.
- 11:07SPEAKER_00: Starting early in one ecosystem means your AI gets smarter faster.
- 11:12SPEAKER_01: Think of it like compound interests.
- 11:15SPEAKER_01: The business that starts using Copilot today is six months ahead of the one that starts in December.
- 11:21SPEAKER_01: Not because Copilot got better, but because Popilot got better at understanding business.
- 11:27SPEAKER_01: The Microsoft OpenAI split isn't drama for drama's sake.
- 11:32SPEAKER_01: It's the most visible signal yet that the AI industry is moving from a cooperative phase to a competitive one.
- 11:39SPEAKER_01: And in a competitive market, the winners are the ones who pick a lane and commit.
- 11:44SPEAKER_00: Let me give you a concrete timeline.
- 11:47SPEAKER_00: August is when Polaris becomes the default.
- 11:49SPEAKER_00: That means right now, June and July, is when your team should be testing.
- 11:54SPEAKER_00: If your developers use Copilot, have them track which suggestions they accept and which they reject this month.
- 12:00SPEAKER_00: Then compare in September when Polaris is live.
- 12:03SPEAKER_00: If acceptance rates drop, you know the new model isn't as good for your code base.
- 12:08SPEAKER_00: That's smart.
- 12:09SPEAKER_01: Don't wait until August to discover there's a problem.
- 12:12SPEAKER_01: Start measuring now so you have a baseline.
- 12:15SPEAKER_00: For non-developers, the same principle applies to the AI features in your Microsoft 365 apps.
- 12:22SPEAKER_00: The AI that summarizes your emails, drafts your documents, analyzes your spreadsheets.
- 12:27SPEAKER_00: All of that is migrating to Microsoft's own models over the next year.
- 12:32SPEAKER_00: Pay attention.
- 12:33SPEAKER_00: If the quality changes, speak up.
- 12:36SPEAKER_00: Microsoft is more likely to fix issues when enterprise customers complain.
- 12:41SPEAKER_01: And here's something most small businesses won't think about.
- 12:44SPEAKER_01: The AI model switch affects your data in ways you can't see.
- 12:47SPEAKER_01: GPT, FOR, and Polaris might interpret the same prompt differently.
- 12:52SPEAKER_01: The summary you get from Copilot and Teams might emphasize different points after the switch.
- 12:57SPEAKER_01: The formula suggestion in Excel might take a different approach.
- 13:01SPEAKER_01: Same input, different output.
- 13:03SPEAKER_01: That's not a bug.
- 13:04SPEAKER_01: It's a model change.
- 13:06SPEAKER_00: Which is why you should never blindly trust AI output.
- 13:09SPEAKER_00: We've said it before on this show and we'll say it again.
- 13:12SPEAKER_00: AI is a drafting tool, not a final product.
- 13:15SPEAKER_00: Read what it gives you.
- 13:17SPEAKER_00: Verify it.
- 13:18SPEAKER_00: Especially after a model change, when the AI's behavior might shift in subtle ways you don't expect.
- 13:24SPEAKER_01: Let me close with this.
- 13:25SPEAKER_01: The AI industry is fragmenting.
- 13:28SPEAKER_01: Every major tech company is building their own stack.
- 13:30SPEAKER_01: That means more choices for businesses, but also more complexity.
- 13:34SPEAKER_01: The businesses that thrive in this environment aren't the ones that chase every new model.
- 13:38SPEAKER_01: They're the ones that pick an ecosystem, go deep, and let the AI compound inside their workflow.
- 13:44SPEAKER_00: Microsoft just made that choice obvious.
- 13:47SPEAKER_00: They picked their own model over open AIs.
- 13:50SPEAKER_00: Now it's your turn to pick yours.
- 13:52SPEAKER_00: That's the show.
- 13:53SPEAKER_00: Control AI profit is back tomorrow.
- 13:56SPEAKER_00: Check what ecosystem you're in, and make sure your AI tools match.
- 14:00SPEAKER_01: And if you're still paying for both, pick one.
- 14:03SPEAKER_01: Your wallet will thank you.
- 14:04SPEAKER_00: Seriously, the amount small businesses waste on overlapping subscriptions would fund a decent AI strategy if they just consolidated.
- 14:12SPEAKER_01: One ecosystem, one AI, one bill.
- 14:15SPEAKER_01: That's the move.
- 14:17SPEAKER_00: See you tomorrow.