A post by Ira Bodnar — founder of Ryze AI — went viral on X. The startup built an AI service that launched and managed ads on Google and Meta: connect your ad account — AI does everything. Hundreds of paying clients in two months, 70% deal conversion. Then Anthropic added direct Meta ad account integration to Claude — and conversion dropped to 20%. Why pay a separate service when the AI you already use can do the same thing?
The thread hit 4M views. Y Combinator head Garry Tan weighed in: Claude will find a replacement for your product either way — whether through MCP or regular search. Hoping the model just won't notice you is naive. David K. added a nuance about enterprise: yes, AI handles more tasks end-to-end, but it's scary to spend real money when you don't know when it'll make a mistake.
The Ryze team didn't shut down — they quickly pivoted to working with large ad agencies managing hundreds of accounts simultaneously. Claude can't reach there yet: too complex processes, too many variables, and human oversight is needed.
Key takeaways:
If your product is just a wrapper over a neural network (AI-Wrapper), you're at risk. The next model update could make your work a free feature.
Those with proprietary data that LLMs don't have survive. Only hardcore deep vertical niches with access to unique client data.
Reliability: AI still makes mistakes, so enterprise needs complex logic and human-in-the-loop. Plus extensive testing and evals.
When everyone can build products, the winner has a recognizable brand and strong distribution.