The Model Isn't the Product Anymore
A founder just raised $18.5 million on the premise that AI has no taste.
Thais Castello Branco built Taste Labs around a simple observation. Frontier models are superhuman at objective work. Code compiles or it doesn't. Math is right or wrong. But her theory is: ask the same model to design something, write something, make something people actually want to look at, and it produces the average of everything it has seen. Her fix is unglamorous and very human: a paid network of tastemakers generating curated preference data, sold to the labs so their models can learn what good looks like. Aesthetics as a data business.
Set that against what the money is saying. On Equity this month, Menlo Ventures partner Matt Murphy walked through Anthropic's run from a $4 billion pre-revenue Series C to a reported $47 billion revenue run rate. His conclusion was not that Menlo picked the best model. It was that the model was never the whole story. The platform around it, the coding tools, the protocols, the things enterprises could actually deploy, is what separated it. His stated lesson for founders: a great model isn't enough, it's the harness that is the difference maker.
The same week, a Chinese open-source model good enough to spook Wall Street made the point from the other direction. When capability is this widely distributed, capability is not a moat. Everyone gets the same engine. The differentiation moves up the stack, to judgment, to product, to knowing what customers want, to discerning the good from great.
The scarce input is no longer access to artificial intelligence. It's taste, which does not scale with GPUs and cannot be raised in a funding round. A solo designer with strong opinions and a $20 subscription now competes on the axis that actually matters. The founders getting funded right now, and Taste Labs is the cleanest example, are the ones selling judgment back into a market drowning in capability.
The takeaway if you're shipping: stop benchmarking your product on what the model can do. Every competitor has the same model. Benchmark on what you ship to your customer.
Sources: This Week in Startups, "Why AI has no taste and how to fix it" with Thais Castello Branco of Taste Labs (July 31); Equity, Matt Murphy of Menlo Ventures on Anthropic's growth (July 22); Equity, "'AI communism', rogue models, and why Kimi K3 spooked Wall Street" (July 24).