Divergent Compute.AI Economic Think Tank

The instrument · second clock

We instrument ourselves to index the future.

Capex Watch reads the financing runway from the outside. This reads the productivity lag from the inside — by measuring, in real time, how much cognitive work AI absorbs inside our own research desk. The firm is the laboratory. The telemetry is the byproduct of doing the work.

The method

Three buckets, not one number

Track every moment of "friction" naively and you measure bad software and human taste, not intelligence. We split it by cause — only one bucket is the signal.

F_env

The Plumbing

Fighting the tools — auth, latency, workarounds. Noise. An engineering problem that drops to zero as the pipes get fixed.

F_cap

The Brain

The AI got the substance wrong and needed correcting. The core signal — when it nears zero, the task is automated.

F_dir

The Steering Wheel

The human changing strategy, tone, taste. Not a failure — the Human Moat. It maps what AI cannot replace.

Sample 001 · the desk measuring itself · 2026-06-25

The readout

A founding-day session: brand, three pages, business model, this instrument. n=1, illustrative — the first sample of the loop running on ourselves.

The Brain · F_cap

0.11

Capability friction per deliverable — near the floor. The AI nailed the hard cognitive work.

↓ lower is better · approaching automated

Autonomy Bonus · SCR

50%

Half the capability errors were caught and fixed by the AI before the human saw them.

↑ higher is better · self-correction

The Human Moat · F_dir

0.55

The human spent energy steering strategy and taste — not fixing basic mistakes.

the value · what AI does not absorb

The Plumbing · F_env

0.67

High drag from clunky tooling — but it never reached the final output.

noise · non-economic · fixable

Illustrative, n=1 — directional, not precision. The shape matters, not the decimals.

The AI is highly capable, the human is acting like a true director, and someone needs to fix the plumbing.

Adoption-frontier sample n = 1 · illustrative Alpha: unproven — calibrating

The slope is the alpha; the level is noise. As samples accumulate at fixed tooling, a durable decay in the Brain — and a persistent Moat — is the leading read on AI's real economic absorption, months ahead of the public statistics. Until that lead is demonstrated against a public outcome, we carry the claim as calibrating, in the open.