In early 2025, Mitchell Jones sat in his basement in Ohio and burned a few hundred dollars trying to build an AI agent that fills out forms, a job that should have taken an afternoon. He kept paying for access to the same models through different wrappers. That friction turned into a company.
Bill talks with Mitchell, founder and CEO of Lava, about how that basement experiment became a business, and then a different business. Lava started as a gateway: one key and one bill for every model, on the bet that AI is a utility and every utility ends up on a meter. Then users at an event kept asking to plug it into Claude. Instead of telling them they were using it wrong, Mitchell asked why. That answer turned Lava into what it is today, an AI agent that works inside the tools a business already runs on, like QuickBooks, Gmail, Salesforce and HubSpot, with no engineering required.
They get into why Cursor's simple layout (your work on the left, an agent on the right) beat the blank chat box, why customers rarely describe their pain cleanly, and why you should fall in love with the mice problem instead of your mousetrap. Mitchell makes the case for planning in three-, six- and nine-month windows while still building an all-weather business, and for building something that gets better as the models get better instead of fighting the frontier labs.
Bill closes with a challenge for leadership teams: if you aren't using AI every hour of the workday, ask your AI what you're missing. Then find the enthusiasts already in your company, buy them lunch every week, and have each one share something cool they did with AI and something they're stuck on.
Before Lava, Mitchell co-founded LendTable, a Y Combinator company he scaled to about 40 people and $40 million raised. He's a Forbes 30 Under 30 honoree and has worked at Meta and spent time at Yale.
In This Episode
- How a basement side project turned into Lava
- Why AI is a utility, and why every utility ends up on a meter
- The customer question that turned a payment gateway into an AI agent
- What Cursor got right that most AI tools still miss
- Love the mice problem, not your mousetrap
- Why founders should plan in three-, six- and nine-month windows right now
- Bill's hourly AI rule and the weekly lunch-and-learn for your flag bearers
- Why you shouldn't trust an AI answer that came back in two seconds
Guest Links
Host Links
Scaling Up Business Podcast with Bill Gallagher
