Key Takeaways

  • Tenex co-founders Alex Lieberman and CJ Hess argue vertical AI startups face extinction as buyers demand custom software built for their exact workflows.
  • Enterprise adoption remains brittle: Tenex finds that only 5% of non-engineering AI agents actually reach production.
  • CJ Hess warns that standalone agent products will collapse unless they control proprietary compliance records or exclusive data sources.
  • Swyx defends vertical software, noting that Wall Street traders pay $50,000 annually for Bloomberg terminals despite identical data existing on public exchanges.

The Disagreement

Alex Lieberman and CJ Hess run Tenex, an AI-native engineering consultancy that builds bespoke systems for businesses. Through that work, they reached a blunt conclusion: vertical AI vendors are in serious trouble.

Lieberman sees enterprise buyers walking away from standard subscriptions because off-the-shelf software never matches their messy internal habits. “There is so much inertia and so many systems you have to wrangle,” Lieberman said. “When a buyer of a vertical solution is expecting truly a one-of-one experience, we are in an age where people are expecting one-of-one software solutions.”

Hess points out that when the core product is merely an automated workflow agent, frontier models eventually eat it from below. “When I think about products where they are really pushing some agent as the product that is doing work in some specific vertical, unless there is some strategic compliance or data moat, I think those are going to struggle,” Hess explained. Companies will either build their own agents in a few afternoons or wait for the base models to handle the job natively.

Swyx pushed back. He argued that human habit and dedicated interfaces protect vertical software from extinction. Software moats do not come from proprietary algorithms; they come from muscle memory and hard-to-reach industry data.

“As long as people train in that software, it never goes away,” Swyx countered. “As a banker in my finance career, I could always access the same exact information from the NASDAQ or the New York Stock Exchange. But why do I pay $50,000 a year for Bloomberg?”

Who's Right (and When They're Wrong)

Lieberman and Hess are right about point solutions. If your vertical AI tool is an automated legal intake form, a medical scribe wrapper, or a generic customer support agent, enterprise buyers will replace you. Today, an internal developer can build a custom agent over a weekend using off-the-shelf foundation models. When standard SaaS requires expensive teams of forward-deployed engineers just to fit an enterprise client's database, the economics break down.

Swyx is right when software serves as the undisputed system of record. Bloomberg does not survive on its intelligence; it survives because every trader knows the keyboard shortcuts and every counterparty uses Bloomberg chat. If an AI startup owns the compliance ledger, the primary workflow interface, and the data inputs that OpenAI cannot scrape, it stays protected.

Vertical startups lose when they sell labor. They win when they sell the desk where that labor happens.

What to Do With This

Run an internal audit of your software product this week. Ask your engineering lead: if a client assigned two developers to build an internal clone of our product using Claude 3.5 Sonnet, what exact feature would stop them? If the only moat is your prompt sequence or UI design, start shifting your roadmap toward proprietary data capture and system-of-record status before internal teams build past you.