Y CombinatorGood News For Startups: Enterprise Is Bad At AI
Harj Taggar on enterprise AI Struggles Give Ambitious Startups A Rare Opening Shot.
In this episode of Y Combinator, featuring Harj Taggar and Jared Friedman, Good News For Startups: Enterprise Is Bad At AI explores enterprise AI Struggles Give Ambitious Startups A Rare Opening Shot The hosts dissect an MIT study on AI project failure rates and argue that its viral, pessimistic interpretation is misleading; instead, it reveals how badly most enterprises execute on AI. They explain that big companies overwhelmingly try to build AI internally or via consultants, and these efforts usually fail due to politics, legacy systems, weak product sense, and engineers who don’t really believe in AI. In contrast, specialized startups that deeply integrate into enterprise workflows and build AI‑native products are winning large deals quickly and decisively. The episode frames this as unprecedented good news for founders: enterprises are desperate, more open than ever to young startups, and switching costs will create strong moats for those who execute well.
At a glance
WHAT IT’S REALLY ABOUT
Enterprise AI Struggles Give Ambitious Startups A Rare Opening Shot
- The hosts dissect an MIT study on AI project failure rates and argue that its viral, pessimistic interpretation is misleading; instead, it reveals how badly most enterprises execute on AI. They explain that big companies overwhelmingly try to build AI internally or via consultants, and these efforts usually fail due to politics, legacy systems, weak product sense, and engineers who don’t really believe in AI. In contrast, specialized startups that deeply integrate into enterprise workflows and build AI‑native products are winning large deals quickly and decisively. The episode frames this as unprecedented good news for founders: enterprises are desperate, more open than ever to young startups, and switching costs will create strong moats for those who execute well.
IDEAS WORTH REMEMBERING
5 ideasMost enterprise AI failures are about execution, not AI being a scam.
The MIT study primarily captures internal and consultant‑led projects, which often fail due to bad software, organizational politics, and weak product execution—not because AI is inherently useless.
Startups that go deep into business processes can massively outperform incumbents.
Companies like Tactile, Greenlight, Castle AI, and Reducto win by embedding into core systems of record, understanding domain workflows, and building AI‑native products rather than shallow ‘AI add‑ons’.
Enterprises’ preference for incumbents and consultants is breaking down under performance pressure.
Banks and FAANGs initially default to trusted vendors like Ernst & Young or legacy software providers, but repeatedly return to startups after those efforts fail to deliver working AI systems.
There is a ‘startup‑shaped hole’ where polymath builders are missing in enterprises.
Successful AI products require rare combinations of cutting‑edge AI knowledge, strong product taste, and deep empathy for human processes—skills that are scarce in large orgs but common in top startup founders.
Winning enterprise AI deals requires navigating politics and cultivating internal champions.
Startups succeed by forming real relationships with risk‑tolerant employees, often people who fantasized about doing a startup, and by leveraging founders whose companies were previously acquired into big firms.
WORDS WORTH SAVING
5 quotesThe majority of software that actually gets built in the world is very, very bad.
— Jared
Apple, a company with infinite resources and infinite access to the smartest people in the world, cannot make a good calendar app.
— Jared
For now, there's just this startup-shaped hole in basically every process or every sort of annoying system that should exist that doesn't exist yet.
— Gary
If your engineers don't believe in this, then how are you gonna build a product that actually works?
— Jared
All these people who are worried that these ChatGPT wrappers won't have moats—like, that's the moat.
— Jared
QUESTIONS ANSWERED IN THIS EPISODE
5 questionsHow can a new AI startup practically ‘embed’ itself into an enterprise’s systems of record without overextending its small team?
The hosts dissect an MIT study on AI project failure rates and argue that its viral, pessimistic interpretation is misleading; instead, it reveals how badly most enterprises execute on AI. They explain that big companies overwhelmingly try to build AI internally or via consultants, and these efforts usually fail due to politics, legacy systems, weak product sense, and engineers who don’t really believe in AI. In contrast, specialized startups that deeply integrate into enterprise workflows and build AI‑native products are winning large deals quickly and decisively. The episode frames this as unprecedented good news for founders: enterprises are desperate, more open than ever to young startups, and switching costs will create strong moats for those who execute well.
What specific skills or experiences best develop the polymath mix of AI expertise, product taste, and domain understanding the hosts describe?
How should an enterprise leader respond if their own engineering org is skeptical of AI but they see clear competitive pressure to adopt it?
What are the ethical and strategic risks of enterprises becoming heavily locked into a single AI vendor due to high switching costs?
For an individual engineer in a big company, what is the most effective way to experiment with AI tools and prove their value internally?
EVERY SPOKEN WORD
Install uListen for AI-powered chat & search across the full episode — Get Full Transcript
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome