Good News for Startups: Enterprise Is Bad at AI (Category AI)

In this podcast episode, Y Combinator’s partners discuss The GenAI Divide: State of AI in Business 2025, which I reviewed, and its widely circulated “95% of enterprise AI projects fail” finding. They explore why this happens and how AI startups might benefit:

• What lies behind the viral “95% failure” statistic?
• Why do internal IT projects stall even at large companies?
• What makes complex enterprise AI systems hard to adopt?
• How do small startups beat banks and consultancies?
• Why do established companies struggle to build working AI products, and what can they do?

The Y Combinator managing partners’ view is roughly this:

• The catchy “95% fail” phrase is misleading: it concerns internal corporate solutions that cannot learn from feedback or adapt. Their argument is that AI itself is not the failure; large companies struggle to build it. They criticise Apple’s calendar app: if even Apple produces poor software, ordinary companies will struggle too. Much enterprise software is mediocre, and expensive consultants such as E&Y or Deloitte often create “two problems instead of one.”
• Large organisations must reconcile many teams’ interests, overcome politics and coordinate requirements. Add legacy systems and design by committee, and AI implementation becomes slow and difficult.
• Startups succeed where corporations struggle: small teams build for specific problems with AI included from the outset and rapidly get to a working result. The speakers cite Tactile, which built a banking AI decision system in months, while Citi and JP Morgan spent years and millions on unsuccessful equivalents.
• They see an ideal moment for startups: large companies want AI badly enough to take risks on young teams. Fear of falling behind speeds decisions and makes buying an existing solution more attractive than building one. Demand exceeds internal capability, leaving opportunities across many niches.
• They also propose a human explanation for failed adoption: some corporate engineers do not believe in AI, avoid the tools and welcome studies that validate their scepticism. In this view, that culture makes failure self-fulfilling and leaves opportunities unused. Companies whose internal teams have given up then call on startups.

A claimed 95% failure rate is not a reason to avoid trying; there is still an opportunity to join the successful 5%. The partners of a major startup incubator predict that corporations will increasingly buy startup products because internal teams cannot deliver :). I would account for that conflict of interest and see an opportunity for corporations to develop their own AI capabilities too, not just an opening for startups.

#AI #Engineering #Software #Leadership #ML