Your AI Platform Is Also Your Competitor and Your Landlord. Anthropic Just Showed Both at Once.
The Information published a piece this month titled "Anthropic Blindsides Its Business Partners." The detail that should stop you: weeks before Anthropic revealed Claude Design (a tool for generating designs and app prototypes), it asked firms including Figma and Canva to be "partners" in the launch announcement. Claude Design competes directly with what Figma and Canva sell. They were invited to help applaud the thing aimed at their own market.
If you build a product on top of a frontier-model API, that's not gossip about big companies. That's a preview of your own risk, written large. The platform you depend on is two things at the same time: a competitor that can ship your feature, and a landlord that can change your rent. Anthropic just demonstrated both inside one quarter.
The competitor problem
Start with the obvious one. When your product is "we wrapped a good model in a nice workflow for designers," the company that makes the model can do that too, and they have the model, the distribution, and a billing relationship with your customers already. Claude Design is the clean example. A whole category of design-and-prototype tools now competes with a first-party feature from the lab whose API many of them were probably calling.
This isn't Anthropic being uniquely predatory. It's structural. Platforms expand into the most valuable adjacent use cases of their platform, every time, because that's where the demand signal is loudest. OpenAI did it to a thousand "summarize my PDF" startups the day it shipped file handling. Apple has done it to indie Mac apps for fifteen years; it even has a nickname, getting "Sherlocked." The new wrinkle is speed. An AI platform can absorb your feature in a release cycle measured in weeks, and announce it at an event it invited you to.
The landlord problem
Now the rent. In April, Anthropic switched Claude Enterprise customers from a flat per-seat fee to usage-based billing with little warning. One compliance consultant quoted in the coverage estimated the change could double or triple costs for heavy users. There was an exemption for businesses paying for fewer than 150 seats, so the smallest accounts were spared, but the direction of travel is what matters. The price of the thing under your product is not fixed, and it can move under you on the provider's schedule, not yours.
Put the two together and you get the real lesson. Building on a platform exposes you to feature risk (they ship your product) and price risk (they re-meter your costs), and a frontier-model platform can pull both levers faster than any platform you've built on before. If your margins assume today's API price and your differentiation assumes the platform won't enter your niche, you've made two bets you don't control.
What I'd actually do about it
I'm not telling you to stop building on these APIs. That's not realistic, and the upside is too good to give up. I'm telling you to build like a tenant who's read the lease, not one who assumes the landlord is family.
Own the customer relationship and the data. If the model provider has a direct billing relationship with your users and you're just a reseller of their tokens, you're the most replaceable link in the chain. Be the layer that holds the account, the history, the integrations, the trust: the stuff that doesn't transfer when they ship a competing feature.
Keep a portability seam. Don't bake one vendor's exact tool-calling format and quirks into your core business logic. A thin abstraction over the model, plus a set of eval prompts you can re-run against a different provider in an afternoon, is cheap insurance against both a price hike and a "we're entering your category" announcement. You may rarely switch. The point is that you can, and the provider knows you can.
Price in a buffer. If a provider can double your unit cost with a billing change, your pricing needs headroom to survive that without going underwater the same week. Margins that only work at the current API rate aren't margins, they're a countdown.
And don't build a business that's one slash-command away from being absorbed. If your entire product is a single obvious capability that the platform will plausibly ship as a built-in, you're not a company, you're a feature in a trench coat. Go where the platform won't follow: deep vertical workflows, messy integrations, the regulated and unglamorous corners where a generalist first-party feature isn't enough.
The honest counter-take
Here's where I'll argue against my own alarm. None of this means Anthropic, or OpenAI, or Google is acting in bad faith, and treating every platform as an enemy will make you slow and paranoid for no payoff. Most of the time, the platform expanding the pie helps you more than its occasional land-grab hurts you: a better, cheaper model lifts every product built on it, including yours. Plenty of durable businesses live happily on top of AWS, which competes with half its own customers, because the value of the platform outweighs the risk of it. You can absolutely build on Anthropic and win.
The move isn't to flee. It's to stop pretending the risk is zero and to spend a little of your time, now while things are calm, making yourself harder to Sherlock and cheaper to re-platform. Build on the API. Just don't let the API own the parts of your business that are supposed to be yours.
Author
Lukas
@lukcombinator