October 17, 2026
Replit CEO Amjad Masad confirmed over the weekend that OpenAI, not Anthropic, is now the platform's default model, according to TechCrunch. Claude isn't gone -- customers can still select it -- but the default is the choice most users never bother to change, and defaults are where real usage lives. This is not a story about which model is 'better.' It's a story about how fragile the foundation under AI-native products actually is.
If you're building a business on top of someone else's AI coding tool, you've just watched that tool's core engine get swapped out from under millions of users with a weekend announcement. That's the risk every company takes when it outsources the thinking layer of its software to a single model provider it doesn't control. It's also exactly the argument for owning more of your own stack rather than renting all of it -- something we've written about in the context of buying vs. building vs. using ViibeStack. Vendor lock-in isn't just a pricing problem anymore; it's a product-stability problem, and Replit's users just got a live demonstration.
In a separate but related signal, Anthropic itself disclosed that its own workforce has gotten less junior over the past year and is spending noticeably more time in meetings, per TechCrunch's reporting. For a company whose entire pitch is that AI automates away repetitive work, that's a telling admission. Rapid growth at AI labs isn't just producing more code and more model releases -- it's producing more coordination overhead, more senior hires to manage complexity, and more of the unglamorous operational friction that every scaling company eventually hits.
The lesson for business readers isn't about Anthropic specifically -- it's about what happens when any organization scales fast on top of AI tooling. More senior headcount and more meetings usually means more process, more approvals, and more of the internal-tooling sprawl that AI labs are supposedly immune to. It's a useful gut-check against the narrative that AI companies are uniquely lean. Even the companies building the frontier models still need the same boring infrastructure -- HR processes, project tracking, internal dashboards -- that everyone else does, which is the exact gap tools like internal tools and admin software and HR and people ops platforms exist to fill.
A platform switching its default AI provider and a lab admitting its own workforce is getting more bureaucratic are both symptoms of an industry moving from experimentation into operations. The novelty phase of generative AI -- pick the shiniest model, ship fast, worry later -- is giving way to a phase where defaults, contracts, and internal process actually matter, because they're what determines whether a tool is still reliable six months from now. Business buyers evaluating any AI-dependent platform should be asking who controls the model underneath it, how easily that could change, and what happens to their workflows if it does.
Actor Mark Wahlberg told TechCrunch he's open to AI being used in filmmaking 'the right way,' including the idea of an AI clone eventually carrying his likeness forward. It's a Hollywood story on its face, but it's really the same question as the other two: who controls the underlying asset once you let AI sit on top of it, and what are the terms when that control changes hands? Wahlberg's comfort with the idea hinges entirely on consent and enhancement rather than replacement -- a distinction that's going to come up in a lot more industries than film before this decade is out.
If Replit can swap its core AI provider over a weekend, how confident are you that the AI tools your business depends on today will look the same a year from now?
Sources