On the Latent Space podcast, Tenex cofounder Alex Lieberman and engineer CJ Hess described how the firm scales its earlier bet on paying AI engineers for output rather than hours. Tenex has grown to 75 people and still uses story points as a base for compensation. The feared failure modes—inflating estimates or shipping bad code to farm points—have not been the biggest problem, Lieberman says, in part because Tenex's AI strategists are responsible for client retention and growth and push back on work that does not serve the customer. The harder problem is rewarding work that does not fit into a ticket, such as understanding client context, designing architecture, and deciding what to build.
Tenex's answer, which Hess demonstrated, is MetaHarness: a shared set of skills, project documents, and a CLI where plans and decisions are checked into Git alongside code. When an agent returns to a task, it can read not only what was built but why the team chose that approach. The useful distinction, the episode argues, is between making agents write code faster and making the surrounding work legible to both humans and agents. The company's own write-up notes that this level of structure can be overkill for a small team on a single project.
The hosts also discussed Kill My SaaS, an experiment in which 72 teams tried to replace a $43,000-a-year speaker-management tool with AI agents. The team chosen by human judges came 19th on the AI judge's leaderboard, a reminder that checking whether software satisfies a list of scenarios and judging whether people want to use it are different jobs. Lieberman cautioned that building software in-house gets riskier upmarket, and someone has to maintain whatever is built. The episode's vertical-AI debate ended on a sharper question: which products become a durable place where people work, and which become just another feature of the agent they already use?