AI for the Fortune 2000: How Tenex Builds an AI-Native SDLC — Alex Lieberman & CJ Hess

What it claims

Tenex is a New York applied AI firm of 75 people, 18 months old as of the interview. Ninety percent of staff are engineers or "AI strategists" (technical consultants). Customers are split evenly between Fortune 2000 companies and mid-market firms with $100M to $500M in revenue. Alex Lieberman says story-point pay has produced roughly a 4x gap between the highest- and lowest-paid engineers, not 10x. He says sandbagging estimates and shipping bad code have not been real problems, because AI strategists are paid on customer retention and growth and push back on unneeded work. The unsolved problem is paying for non-coding work such as architecture and client meetings. CJ Hess says Tenex recalibrates story-point criteria at least once a quarter, since the roughly 2x output gain should in theory deflate a point's value.

How it works

The MetaHarness has three parts: a standardized skill set, a structured documentation tree inside each repo, and a CLI that the models use to produce plans, visualizations and client readouts. Each project has one canonical plan document. CJ says Tenex has loosened the plan's enforced structure as models improved. Agents follow an opinionated convention for logging their decisions and progress, and everything is committed to Git. CJ prefers these outputs as HTML "picture books" rather than Markdown. His stated rationale is that code alone cannot record why a decision was made. Tenex measures whether the process works by tracking average story points per engineer per month.

For tooling, CJ says most of the team drives with Opus 5.5, uses Codex as a reviewer working back and forth on the same machine, and uses Codex for computer-use tasks. He runs Opus through AMP's cloud "orbs" with a Claude Code subscription attached, which the host says cuts cost by about 80%. CJ says he could not reproduce his cross-model review workflow in Claude Code in the cloud.

Evidence

All figures are self-reported in conversation, with no method given:

• Output: about 2x per engineer over 2026, counted from the Opus 4.5 release in December 2025.
• Token spend: roughly 1:1 to 2:1 against employee spend. The host's framing puts typical companies at $10 to $200 per employee and AI startups at about 1x. Costs fell after Tenex moved from Astra and Fable to Opus 5.5.
• Enterprise adoption: Lieberman's figures come from about 100 enterprises Tenex is talking to:

• Kill My SaaS: Latent Space was quoted $43K a year in the post, or $40K in the transcript, and spent $1,000 replacing the tool. There were 72 submissions. The human judges' winner placed 19th on the AI judge's leaderboard.

Common customer asks are back-office agents and data permissioning (RBAC) for employees using Claude or Codex. Tenex first built bespoke gateways and now sells its own gateway product.

Limits

Tenex's own SDLC article says the MetaHarness structure can be overkill for a small team on a single project. CJ says agent deployments fail when no standard process exists: when 40 people do one function 40 ways, evals cannot be built and an agent cannot tell whether it is right. Lieberman argues that replacing upmarket systems such as Salesforce or SAP is too risky, and that any in-house tool carries ongoing maintenance load. The transcript is auto-generated with unlabeled speakers, so some attributions are inferred from context. The page links to, but does not reproduce, the MetaHarness file layout and CLI commands.