AI for the Fortune 2000: How Tenex Builds an AI-Native SDLC — Alex Lieberman & CJ Hess
Latent Space
Tenex says AI engineers should be paid for output, not hours. Now it’s trying to make that work inside a 75-person team serving Fortune 2000 clients.
Based on reporting by Latent Space — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Last year, Alex Lieberman made a blunt bet: if AI makes engineers faster, paying by the hour starts punishing the people doing the best work. On his return to AIE New York, he and CJ Hess were there to explain what happens when that idea stops being a theory and becomes a real company policy, with customers and payroll attached.
Tenex still ties compensation to story points, but the company says the scary parts of that model have not been the main issue. The bigger problem is more human and more annoying: how do you pay someone for the work that isn’t a ticket? That includes architecture, client conversations, and the hours spent turning a customer’s messy context into something a model can actually use. As CJ put it, the job is shifting from typing code to supplying the model with the right input.
That shift matters because Tenex is no longer a tiny experiment. Alex says the company now has 75 people, and the customer mix is split between Fortune 2000 firms and mid-market businesses. It started as an AI-powered development shop, but customers are increasingly asking for help with AI transformation across the company, not just engineering support. That broadens the ask, and it also makes the compensation question messier.
Tenex’s answer is MetaHarness, a shared workflow built around documentation, skills, and a CLI. Plans and decisions live in the repo alongside the code, so when an agent comes back to a task it can see not only what was built, but why it was built that way. The point is less “make the model code faster” and more “make the whole job legible,” including acceptance criteria, evidence, and the logic behind decisions. Tenex also admits that this kind of structure can be overkill for a small team on one project. The company is basically building process so the process can be read by machines.
The conversation also circled back to Kill My SaaS, where 72 submissions tried to replace a speaker-management tool priced at $43,000 a year. The team’s AI judge and human judges picked different winners, which is the whole story in miniature: software can satisfy a checklist and still be the wrong thing to live with. Alex warned that as companies get larger, ripping out an existing system gets riskier and more expensive. But the temptation to own the workflow is obvious. Better tools are nice; owning the thing you hate is the real lure.
My take — AI-written commentary, not fact-checked reporting
The real story here isn’t “AI engineers get paid differently.” It’s that companies are starting to price judgment, context, and maintenance instead of lines of code, which is much less romantic and much closer to reality. Everyone loves the idea of AI-native work until someone has to own the repo, the client, and the mess.
Read more about this at: Latent Space