Ripping out mortgage cores: Vesta's Mike Yu on software for AI agents
Key points
Key takeaways from the Source podcast interview with Vesta co-founder and CEO Mike Yu (October 2026):
From Jane Street to mortgages. After interning as a trader at Jane Street, Yu persuaded Blend to take him on as its first ever product intern, grinding through roughly 3,000 accessibility defect findings to get Wells Fargo over WCAG 2.0 AA compliance and later working with Fannie Mae and Freddie Mac on data verification programmes.
Four years at Blend. Yu spent four years at Blend, where he met co-founder Devon Kumar, the company's first engineer, building an insurance call centre and learning the mortgage ecosystem before founding Vesta in 2020.
Replacing the system of record. Vesta replaces lenders' legacy core systems with a single system of record covering compliance, rules, workflow and integrations, and its AI agents take over the back-office work that today keeps lenders' staff reading documents for around 45 days per loan.
Designed so software can do the work. Yu argues Blend showed the ceiling of what software layered on top of legacy cores can achieve, so Vesta rebuilt the system so each task can be performed by a human or by software interchangeably.
An early bet on document intelligence. Before LLMs, the computer-vision era of OCR and document extraction convinced Yu that classical rules plus extraction could automate roughly 80% of mortgage origination work — a conviction baked into Vesta's architecture.
Why US mortgages cost so much. Yu points to America's 30-year fixed-rate mortgage — a bet no other country made — as the structural reason underwriting is so hard: a lender must underwrite three decades of fixed payments, which keeps back-office costs high and automation attractive.
Automation is proving out. Customer Pennymac has reached a 25% automation rate on mortgage workflows, and Yu sees lenders' willingness to move mission-critical operations onto Vesta as the signal the AI-native system of record works.
Why AGI does not make Vesta obsolete. Yu argues the hard part is not reading documents but replacing the mission-critical core of a heavily regulated industry, which is why he believes AI makes deep vertical software like Vesta more valuable, not less.
Read more: a16z on YouTube
