A $650M AI founder: Your ‘demo-level’ AI is a ‘real danger’

This article summarizes an episode of Y Combinator’s video series featuring Jake Heller, co-founder and CEO of Casetext.

Jake Heller, co-founder and CEO of Casetext / Photo credit: The Social Radar

Jake Heller is the co-founder and CEO of Casetext, the AI legal startup acquired by Thomson Reuters for US$650 million. With so much AI hype, he believes most founders are missing the point entirely. His approach to building an AI product is to find the best expert in a field, break down their process, and then build a product that automates those exact steps.

A method for finding good AI ideas

Instead of thinking about future ideas, Heller suggests founders should focus on jobs that companies already pay people to do. This makes it easier to find a market for a product and shows three ways to use AI to create something useful.

• Assist. AI can help workers get better at their jobs. This means creating tools that help with their current work, not tools that change it.
• Replace. AI can completely take over a job that people do now. This lets a company compete with other companies that offer the service.
• Do the unthinkable. AI makes it possible to do things that were not possible before because they were too expensive or difficult. This creates new markets and allows businesses to do new things.

A new way to measure market size
Most software companies measure their market by looking at how much companies spend on software. Heller argues that AI companies are not just trying to get a piece of that money.

Heller explains, “The actual amount of money that we already know people and companies are willing to spend is the combined salaries of all the people they’re currently paying to do the job. And that number is like a thousand times bigger.”

From how experts work to working code

Before writing any code, Heller says founders must understand how the best experts do a certain job. This means breaking down how they work into simple and repeatable steps.

The “unlimited resources” test
To find a starting point, Heller suggests a simple question to help founders build their product.

“Ask yourself this question,” Heller advises. “How would the best person in that field do this if they had like unlimited time and unlimited resources… How would the best person do this and work backwards from there?”

Turning knowledge into prompts
Once you know how an expert works, the next step is to turn each step into an instruction the AI can understand. This turns an idea into a technical plan.

Heller states, “Most of these steps for the kinds of things you’ll be doing end up being prompts. One or many prompts… The reason why that many of them are prompts is because they’re the kinds of things that would once require human-level intelligence.”

Building a product through hard testing

After planning the work, founders start the most important and ignored part of development. Heller describes a careful process for making sure AI tools work well enough for real work. He warns that this is where most startups get stuck with a demo that looks good but does not work.

The dangers of “demo-level” AI
To show progress quickly, many teams build demos that look good but only work in perfect situations and fail with real use. This gives a false appearance of success that can misguide a company.

A good product as the main way to sell

More than just software and early sales