The Moat Was the Plumbing

Ira Bodnar, founder of an AI ad management startup, on how Claude's new MCP connectors for Meta Ads cratered their close rate from 70% to 20%:

We built an AI agent that automates ad management for you. Like, you can give it access to all your Google and Meta accounts and it will manage it for you. Pretty cool. Customers loved it.

Customers loved it right up until the integration became a platform feature. A 70-to-20 close rate collapse tells you exactly what people were paying for: the plumbing, not the intelligence. The moment the plumbing got commoditised, the cheque-writing stopped.

This part is real, and anyone building "AI + API access to someone else's platform" should read it twice. Your product is one partnership announcement away from becoming a demo.

Then the post does the thing founder posts always do — it takes a genuine wound and inflates it into prophecy. We get "ads inside LLMs are the next trillion-dollar channel" (the entire value of an AI assistant collapses the moment users suspect it's shilling for advertisers), "every marketer will tell Claude to launch a million-dollar campaign" (as if the hard part of ads was ever clicking buttons in Meta's UI), and "98% of TikTok will be fake AI content" (unfalsifiable, conveniently dramatic, and ignoring that platforms have a $200 billion annual incentive to prevent exactly this).

The most revealing line comes near the end, after 800 words of existential alarm: "Our current business will be fine anyway." Turns out they're pivoting to complex enterprise workflows and agency services — moving from the commodity layer to the complexity layer. That's the actual insight. It's also the opposite of the headline.

The lesson isn't that Claude killed a startup. It's that a startup discovered its value proposition was a feature, not a product. That's not an AI story. That's the oldest story in software.