CharlieKey takeaways from your favorite podcasts
← 21 October 2026Latent Space

AI in NYC: How Tenex Builds an AI-Native Engineering Team — Alex Lieberman & CJ Hess

Tenex pays engineers by output and builds around AI-native workflows, while its leaders argue enterprise adoption lags far behind raw model capability and that standardized processes must come before agents.

Watch on YouTube ↗

Charlie's summary

  1. Pay engineers for output, not time: AI is a leveraged bet on a person's intellect and work ethic

    CJ joined Tenex because he was getting faster with AI yet getting paid less while billing hourly. Tenex pays by story points, so earnings vary more than ever. Alex says the gap between its highest and lowest paid engineers is about 4x, not 10x. The feared problems, sandbagged estimates and sloppy code, have not materialized, because AI strategists are incentivized on retention and growth. The hard part is paying for non-coding time like architecting and client meetings.

    ▶ 1:28
  2. The meta harness: keep every plan and decision in the repo so agents hold the context

    CJ describes a structured set of docs in the repo plus a CLI that models use to drive planning, visualizations, and client readouts. There is a canonical plan document, and agents log their decisions, all checked into Git. As models improved, Tenex loosened the enforced structure. CJ argues code alone is not enough memory: you need to know why a decision was made. Context lives “in the repo”, never in someone's head.

    ▶ 7:07
  3. Engineer output roughly doubled this year, so story points must be recalibrated like inflation

    CJ sees output per engineer about 2x higher since the Opus 4.5 moment in December, when models began handling long-running tasks. Tenex tracks average story points per engineer per month as its test of whether processes work. In theory, points should deflate as models improve, so at least once a quarter the team re-examines its criteria for estimating work, balancing rewarding output against the value being created.

    ▶ 9:39
  4. Token costs are falling per task, so token-maxing is now a smaller risk than it was six to eight months ago

    Alex says Tenex spends roughly one to two times employee cost on tokens. He argues costs per task keep dropping, so it is harder now to wreck the budget by token-maxing than it was six to eight months ago. CJ adds that cheap frontier-level models are arriving, and most economically valuable tasks can be done with Opus 5.5. Would a model ten times smarter justify the cost of moving cells in a spreadsheet? Probably not.

    ▶ 16:44

Give your email to unlock the rest of the key points — for every episode.

Do more with Charlie

Free — sign in with your email, no password.