Anthropic: We're not trying to eat startups' lunch

I spent most of my career hearing big tech firms tell me they were building a platform for my success, only to watch them build a feature that killed my business two years later. It is the natural cycle of the valley. So when Anthropic, the creators of Claude and the current darling of the developer community, hosts a town hall in London to tell founders they aren't trying to eat their lunch, my first instinct is to check if my sandwich is still in the fridge.

Jack Clark, one of the co-founders, recently sat down with a room full of European entrepreneurs to deliver a specific message: Anthropic wants to be the infrastructure, not the application. In his view, there is a clear line between the frontier model and the specialized tools builders are creating on top of it. It is a nice sentiment. It is also one that every platform company in history has expressed right up until the moment they needed to show quarterly growth to their investors.

The Infrastructure Illusion

The core of Anthropic's argument is that building a foundation model like Claude is so expensive and requires such a specific set of research skills that they don't have the bandwidth to build niche vertical software. They want to provide the intelligence, and they want you to provide the interface, the workflow, and the customer relationship.

For a founder, this sounds like a green light. If Anthropic isn't going to build a legal discovery tool or a medical charting app, then you can safely build those things using Claude as your engine. But here is the problem: the line between "infrastructure" and "application" is moving every single week. When Anthropic released "Artifacts" or their new "Computer Use" capabilities, they weren't just updating a model; they were building user experiences that previously belonged to third-party startups.

What Clark is describing is a world where the model is a commodity utility, like electricity. But electricity doesn't suddenly decide it also wants to be a toaster. AI models are different because they are inherently general-purpose. Every time the model gets smarter, it naturally absorbs the functionality of the wrappers built around it.

The Founder's Dilemma

If you are building in the AI space right now, you are essentially playing a game of chicken with the lab that provides your API. You have to move faster than their research cycle. Anthropic’s pitch in London was focused on the idea that they are "safety-first" and "builder-friendly," contrasting themselves with the more aggressive, product-heavy approach of OpenAI.

But we have to look at the math. Anthropic has raised billions of dollars. You do not raise that kind of capital just to be a silent pipe. At some point, the pressure to capture more of the value chain becomes irresistible. If they see a specific use case—say, enterprise coding assistants—generating billions in revenue for their API customers, they would be derelict in their duty to their shareholders not to build a first-party version of that tool.

This isn't me saying Anthropic is evil. It's me saying they are a corporation. When a founder hears "we aren't your competitors," they should translate that to "we aren't your competitors yet."

Where the Moat Actually Lives

So, where does a builder go from here? If the platform provider says they won't compete, but the technology dictates that they eventually will, how do you protect yourself? During the London session, there was a lot of talk about the "unbundling" of these models. The idea is that while a general model is good at everything, it isn't excellent at the specific, messy nuances of a particular industry.

The takeaway for founders is clear: your moat cannot be the AI itself. If your value proposition is just "we have a better prompt for Claude," you are already dead. Anthropic will eventually bake that prompt's logic into the model itself. Your moat has to be the things that a research lab in San Francisco doesn't want to deal with.

• Proprietary Data: Stuff that isn't on the open internet for them to train on.
• Workflow Integration: Being so deeply embedded in a user's daily habits that switching to a generic tool is painful.
• Regulatory Expertise: Handling the red tape that a general-purpose AI company won't touch.
• Human-in-the-loop: Providing the service and accountability that software alone cannot offer.

The London Context

It is interesting that Anthropic chose London for this charm offensive. Europe has traditionally been more skeptical of big tech hegemony and more focused on regulation. By positioning themselves as the "partner" to the European startup scene, Anthropic is trying to win the hearts and minds of the builders who are currently annoyed with the closed-door nature of other labs.

They are leaning into the "pro-developer" narrative. They want to be the platform of choice for the sophisticated builder who values stability over hype. And to be fair, their technical documentation and API reliability have been stellar. But founders shouldn't let a good API distract them from the long-term strategic reality.

Final Takeaway for Builders

Anthropic is currently the most honest of the big labs, but honesty in business is often a function of current market position rather than permanent character. Right now, they need you. They need your API fees, they need your feedback to improve the model, and they need you to prove out use cases so they can see what works.

Build on Claude because it is currently the best tool for many tasks. Take their support and their credits. But never stop building the parts of your business that they can't replicate with a model update. The moment you rely on their promise not to compete is the moment you stop innovating on your own defensibility.

Don't build a house on someone else's land and be surprised when they decide to start charging rent—or move into the master bedroom.

Enjoy the lunch they aren't eating today, but keep an eye on the menu for tomorrow.

Read the original at Sifted →