How Tenex Builds an AI-Native SDLC

Summary

On the Latent Space podcast, Tenex co-founder Alex Lieberman and Engineering Director CJ Hess explain how the company structures its AI-native software development lifecycle. Engineers are paid based on delivered outputs rather than hours; plans and decisions are committed to Git via 'MetaHarness'. The company also provides consulting for AI transformation to Fortune 2000 clients. The conversation touches on the 'Kill My SaaS' experiment, which replaced a $43,000 annual agency, and discusses whether vertical AI products can survive as frontier models take over their core tasks.

Why it matters

Tenex's outcome-based pricing model directly challenges the hourly billing structure that dominates corporate software consulting. This serves as a useful stress test for how AI is changing the economics of engineering work. The MetaHarness concept, which treats architectural rationale and acceptance criteria as first-class repository artifacts, offers a concrete template for keeping agent-assisted work readable for both humans and models. Most teams are still solving this problem through trial and error. The vertical AI discussion frames a real strategic rift: are domain tools becoming permanent workspaces, or are they being absorbed as features of general-purpose agents?

Highlights

• Tenex pays engineers based on story points tied to delivered work. The feared manipulation (inflated estimates, low-quality code) has not materialized according to reports. This is supported by AI strategists taking on user retention responsibility as a natural balancing factor.
• MetaHarness commits plans, decisions, and acceptance criteria to Git alongside the code. This allows returning agents to read not just what was done, but why it was done.
• The 'Kill My SaaS' winner (a project ranked 19th by an AI judge) demonstrates that meeting a list of scenarios and building something people actually want to use are fundamentally different evaluation problems.

Critical take

The episode reads more like a confident company narrative than a rigorous case study. Tenex's 75-person scale and Fortune 2000 client list are claimed but not supported by metrics. Given the company's youth, dismissing the risks of billing manipulation feels weak. The vertical AI discussion is entertaining but fails to resolve the underlying tension. No concrete pricing, churn, or adoption data is provided to show whether 'workflow glue' products actually retain users when underlying model capabilities become cheaper.