Garry Tan says he rebuilt his 2008 startup as about 70,000 lines of code in roughly 90 hours, a task he could not do with $5m and five engineers over two years.
YC is letting spring batch applicants upload Codex or Claude Code transcripts, with Garry arguing that how someone prompts agents reveals whether they can build and have taste.
Garry says SaaS is dead unless companies embrace Claude Code and an agentic approach, while the partners see systems of record like Rippling as safer than integration-heavy tools.
YC partners say AI coding makes taste and agency scarce
AI coding agents have made building so cheap that the scarce things are taste, agency and a product people want. YC is adapting by judging founders on how they work with agents and by scaling to four batches a year. The partners see software built on thin integrations as exposed, while marketplaces and systems of record look safer.
Takeaways
AI coding tools collapsed the cost of building. Garry Tan rebuilt his 2008 startup as about 70,000 lines of code in roughly 90 hours, while holding a full-time job. He says he created in 80 hours what $5m and five engineers could not build in two years. He credits Opus 4.5.
YC will now read founders' agent transcripts. Spring applicants can upload Codex or Claude Code session transcripts, run in a security sandbox. Garry argues prompting style shows whether someone can build and has taste. He wants to fund founders who understand the user and are tenacious rather than by school.
The bar for a first product keeps rising. Harj Taggar says the bar for demos a few weeks into a batch has risen every six months for three years. Jared Friedman says pivoting founders with no prior on good ideas launch many things and usually fail.
Competition matters less than making something wanted. Jared Friedman says YC mostly tells founders not to worry about competitors, citing Legora launching despite Harvey. Garry Tan returns to the motto of making something people want. Jack Altman adds that much product-market fit comes from tailwinds outside the startup.
Some software categories look safer from AI. Garry says SaaS is dead unless companies embrace Claude Code and an agentic approach. The partners see marketplaces such as Airbnb and DoorDash as safe, systems of record such as Rippling as safer than integration-heavy tools, and Salesforce as probably in trouble.
Small teams reach scale while capital concentrates. Harj says reaching $1-2m in ARR without hiring is easier now, while Series B rounds are bigger than ever and dollars sit in fewer firms. Garry says capital as a bludgeon works less well when a company no longer needs a thousand people.
YC is scaling to find more great founders. YC moved from two batches a year to four, with each partner running a pod of about 30 companies. It is reaching over 30 college campuses and pushing toward grad students in their mid to late twenties. Jared says the bottleneck is more great founders.
What comes next for YC and venture
Garry Tan expects YC alumni to fill most of the Midas list within 10 to 20 years.
He says venture capital could grow between 2x and 10x in ten years, with most investors only B or B+.
Tap a name to go there · tap a line to see why they are linked
On this episode 4
Show 1
Harj Taggar predicts the next Salesforce will come from a YC batch selling to YC startups.
Paul Graham's view, recounted by Harj, is that YC should produce more companies reaching $100m in revenue.