Tenex pays its engineers for what they ship, not for the hours they work. About a year and a half after the company started, co-founder Alex Lieberman says the experiment has produced a wide spread in pay, though not as wide as the company's name suggests. Its highest-paid engineer earns "probably something around 4X" what its lowest-paid engineer earns, he said. "Not 10X difference yet. That's the dream."
Lieberman and CJ Hess, Tenex's first hire, discussed the pay model with swyx at AI Engineer New York in a Latent Space episode published October 21, 2026. The conversation also brought up a newer problem for the model. If AI keeps making engineers faster, the company has to keep revaluing the unit of work it pays for.
Why hourly billing stopped making sense
Asked by name what drew him to the company, Hess said he had "a graveyard of startups" behind him, ending with his own engineering consultancy. He billed his clients by the hour. As he moved into AI consulting and built more with AI tools, he said, he felt his output getting "better and faster and ultimately getting paid less."
Then he saw a post Lieberman had put on Twitter about rewarding engineers for their output instead of their time. Hess said the advantage wasn't obvious back then, in the days of Anthropic's Claude Sonnet 3.5 and Google's Gemini 2.5 Pro. He bet that his skills, combined with heavier use of AI, would raise both his output and his earning power. Two years on, he said, engineers can run many pieces of work in parallel. He expects the rest of the industry to follow: when AI is solving the technical problems, he said, companies will pay for outcomes and won't want to pay for the time it takes to build something.
How the pay model works
The unit Tenex uses is the story point, a measure software teams commonly use to estimate how big a task is. According to Tenex's description of its engineering service, the company breaks client requests into tasks and estimates them in points. Clients choose priorities in two-week cycles called sprints and are charged for the output produced, and engineers are paid for the points they complete. In the conversation, the arrangement was summed up this way: the points have to be agreed on, and some engineers can make millions of dollars a year while less productive ones make much less.
Lieberman described the reasoning behind it: "AI is really just a leveraged bet on a human's intellect and work ethic." If that is true, he said, earning capacity will vary more than ever. He acknowledged the model "isn't for everyone." It is built for engineers with a strong software foundation who keep using the newest tools to stretch their time while keeping quality high, people who want to "go long themselves," as he put it.
A 4x spread, not 10x
Lieberman gave some numbers on the company. Tenex began about 18 months earlier as an AI-powered development shop. The idea came partly from his co-founder Arman Hezarkhani, who was using coding agents heavily while engineers he knew were still typing code into their editors without adopting the technology at all. Today the company has 75 people, and Lieberman said about 90% of them are engineers or AI strategists, whom he described as technical consultants. Half its customers are Fortune 2000 companies, he said, and the other half are mid-market firms with $100 million to $500 million in revenue.
Asked for an update on how many people earn $1 million or more a year, Lieberman didn't give a count. He answered with the 4x spread instead. In the founders' earlier Latent Space interview from November 19, 2025, seven-figure cash pay for several engineers was presented as something they expected within the following year.
The known weak spots
Lieberman said the company had run into "limits to the model that we knew would be limits." He listed the classic objections to paying by output: story points are subjective, and people may either sandbag their estimates or "ship a bunch of bad codes." The conversation added another risk, that people might barely check the box without really doing the job. Lieberman agreed, but his answer breaks off before he explains how Tenex guards against these failures. In the 2025 interview, the founders had described technical strategists, rewarded on whether customers stay and expand their contracts, as a quality check on plans before they reach clients.
The talk then turned to how engineering work itself has shifted at Tenex. Given the right context, the argument went, today's models can get about 90% of the way to a solution, so more of the value lies in sitting down with the customer, capturing everything they know and passing that to the model. In that view, Tenex may come to judge the delivered output more directly (what it is, how it works, how it looks) while keeping story points as the unit of account.
When the same work takes half the effort
Asked how engineers' output compares with the start of the year, Lieberman said he didn't know. The estimate that followed in the conversation was roughly a 2x increase this year in what engineers can cover in the same amount of time, offered as a personal impression rather than a measurement. It was dated from December, described as the moment Claude Opus 4.5 let models take on long-running tasks, to today's Opus 5.5, praised as a model that can do a great deal of work. Anthropic announced Opus 5.5 on September 22, 2026, and pitches it for extended coding work such as migrations and audits. The gain depends on the person: some Tenex engineers were likened to a picture of a wizard over an orb, directing agents around the sky.
That raised the obvious question: is a story point worth as much now as it was a year ago? Tenex is always trying to calibrate that, the discussion went, and it is a tricky equation. Lieberman said "in theory, there should be deflation," meaning each point should stand for more work than before.
In practice, Tenex sits down at least once a quarter to review its criteria for story points and how it estimates projects. The aim is to keep rewarding output while adjusting for the value the work creates as models improve, and the roughly 2x gain has fed into those estimates in a similar way. Set the features engineers are building and shipping now beside what they shipped at this time last year, the Tenex account went, and the difference is enormous.