Vertical AI Agents Could Be 10X Bigger Than SaaS

Harge compares AI agents with the SaaS boom that followed the spread of XMLHttpRequest, which let websites behave more like desktop apps. He says over 40% of venture capital went to SaaS over the prior 20 years, producing more than 300 SaaS unicorns. Broad consumer categories mostly went to incumbents, while specialized B2B software created room for many startups.

The hosts discuss whether LLMs will let companies grow revenue without hiring at the same pace. Harge suggests hiring engineers who can automate specific growth bottlenecks, rather than building large teams for functions such as customer success or sales. He points to the possibility of a future unicorn operating with around 10 employees.

Harge argues that a vertical AI company can combine software with the people who operate it, potentially capturing payroll spending as well as software budgets. Examples include Outset applying LLMs to surveys, Matic automating QA testing, and A Priora handling technical and initial recruiter screening. The hosts note that selling to teams whose jobs may be replaced can create resistance, so some products need executive-level support.

Harge says simple customer support bots often cannot replace teams handling complex work. He cites Giga ML, which handles 30,000 daily tickets for Zepto and uses an evaluation set of 10,000 test cases. The examples suggest agents may need specialized workflows and data for each industry before broader platforms can serve many customers.

The hosts discuss voice agents in auto loan collections and the rapid improvement in voice quality and latency. They close by advising founders to look for boring, repetitive administrative tasks in fields they understand or can access. Examples include bidding on government contracts and processing dental insurance claims.