What Happens When You Build Software for Agents
a16z Deep Dives · 2026-10-08 · 64 min
https://www.youtube.com/watch?v=TyWe_QQtYEUVideo summary
Vesta CEO Mike Yu says mortgage AI needs a new system of record, targeting 75% automation in a heavily regulated industry.
On the a16z podcast, Kimberly Tan interviews Vesta CEO Mike Yu about replacing mortgage lenders’ legacy systems rather than layering AI onto them. Yu traces his path from Jane Street to Blend, where he learned the industry, then explains how Vesta handles compliance, workflows, integrations, and AI agents. He argues that legacy flat-file databases—with tens of thousands of fields and single-writer limits—block automation, while Vesta’s redesigned system can let agents work in parallel. Mortgage automation rose in 2025, and Yu says Pennymac reported removing 25% of operating costs before Vesta deployed its agents; he estimates the product is about 75% of the way toward automating possible tasks. He also makes the case that long-term rates will stay structurally higher, defends vertical SaaS, and says lender trust, regulation, and implementation expertise will keep Vesta valuable even as AI advances.
Chapters
- 0:00From Jane Street to mortgages: Stanford student Mike Yu joins Blend
- 3:31From Jane Street to mortgages: Yu joins Blend as online applications take off
- 6:20From Jane Street to mortgages: Wells Fargo’s 3,000 accessibility findings and meeting Devin
- 10:15What Vesta actually does: Replacing manual mortgage back offices with AI agents
- 14:06Why rip out the whole system of record: 30,000-field databases block parallel AI agents
- 18:41The case for structurally higher rates: debt, AI CapEx, and a conservative Vesta plan
- 24:28Is the SaaS apocalypse real? Vesta’s vertical software model and the value beyond token pricing
- 30:42Becoming AI-native with Devin: Sonnet 3.6 and event-triggered mortgage agents at Vesta
- 37:46Automation rates and PennyMac’s 25%: Vesta estimates 75% of mortgage work is automatable
- 41:17Why mortgages are so regulated: 30-year fixed loans bring duration risk and extensive government rules
- 44:54Why mortgages are so regulated: state-by-state rules and automated underwriting constrain AI
- 47:04Why mortgages are so regulated: Vesta’s 2022 launch required Fannie Mae and county integrations
- 51:03Why mortgages are so regulated: partners and lenders bet on Vesta’s undocumented LOS expertise
- 53:10Why Vesta survives AGI: lender transformation and 30-year mortgages supporting the middle class
- 56:46Why Vesta survives AGI: a 95-person remote team brings AI to mortgage lenders
- 59:15Why Vesta survives AGI: 12x revenue growth, a 5x pipeline, and product judgment
This is a Tier 1 public summary
Whether the chapter key points, section summaries and mind map are public is up to the person who shared it. Want the full analysis?Submit one yourself.
More from this channel
How Valon Is Rebuilding Mortgage Servicing for the AI Eraa16z Deep DivesValon rebuilt mortgage servicing for a $13 trillion market, reaching 3× efficiency and preparing to transfer 4 million loans with AI.
To Regulate AI Effectively, Focus on How It’s Useda16z Deep DivesMartin Casado argues AI laws should target harmful uses, while regulatory uncertainty pushes U.S. startups toward Chinese open-source models.
Inferact: Building the Infrastructure That Runs Modern AIa16z Deep DivesInferact founders explain how vLLM serves 400,000–500,000 GPUs and why they’re building an open-source inference layer.
How Palantir Scaled: Why the Best Software Is Built Backwardsa16z Deep DivesPalantir Chief Architect Akshay Krishnaswamy explains how forward-deployed engineers turn customer pain into reusable products.
Mintlify and the Transition From Human Docs to Agent Infrastructurea16z Deep DivesMintlify cofounder Han Wang says docs are becoming AI infrastructure, with 20 million people visiting Mintlify-powered sites last month.
Related analyses
Why AI Is Reinventing How Businesses Buy Everythinga16zLio CEO Vladimir Keil says AI agents can manage procurement end to end, where one missed aircraft-part email could cost hundreds of millions.
The Current State of Consumer AIa16za16z finds only 4.5% of US consumers pay for AI, while OpenAI’s ads reach a $1 billion annual run rate.
How AI Will Rewire Consumer Internet | Deep Dives with a16za16z Deep DivesJosh Elman joins a16z after Apple, arguing consumer AI must deliver 10x-better experiences and lasting retention.
Cybersecurity in the Agentic Era | Deep Dives with a16za16z Deep Divesa16z, Neo, and Cotool warn that 50% of enterprise apps could be agentic by year-end, making traditional security defenses unreliable.
Why Top Founders Are Racing Into AI Infrastructurea16za16z’s Machine Age Fund targets AI infrastructure as GPU supply sells out through 2028 and memory demand needs three years of capacity.