Skills diagnostics & practice

Every skill you have is being re-priced.

Skills Infinity measures which parts of your work AI now does outright, which parts you'll be paid to supervise, and which parts compound. Then it trains the third kind.

Skill half-life index rev. Jul 2026
Composition of a working day half-life 2.4 yrs
  • Done by the modelReconciliation, variance tables, first-pass commentary. 46%
  • You supervise itChecking assumptions, catching the plausible-but-wrong. 31%
  • Compounds with youDeciding what to measure and defending it to a board. 23%

Half-life is how long before half of today's task list moves left. Illustrative model — your audit uses your own task log.

The premise

Nobody loses a whole job. They lose a column.

Skills don't disappear on a schedule — they get hollowed out from the inside. The routine middle of a craft goes first, quietly, while the title stays the same. What's left is thinner and harder: framing the problem, judging output you didn't produce, carrying responsibility for a decision a model can't be accountable for.

Generic courses teach the middle column — the part with the shortest remaining life. We start by finding the edges of your own work, then drill the parts that get more valuable each year rather than less.

How it works

Three weeks to a defensible plan.

Week one — Audit

Map your actual tasks

You log a fortnight of real work, or connect a calendar and ticket queue. Each task is scored against what current models do unassisted. Output: your band, not a persona's.

Week two — Drill

Practice on live problems

Graded exercises in the compounding column: judging model output under time pressure, specifying work, spotting confident errors in your own domain. Reviewed by a practitioner.

Week three — Proof

Ship something you can show

You finish with a worked artifact and a scored record of the judgment calls behind it — the evidence a certificate was never able to give a hiring manager.

Tracks

Six starts, built on the compounding column.

Judgment

Reviewing what you didn't write

How to audit model output fast: sampling strategy, error signatures, when to stop trusting fluency.

Specification

Writing the brief

Turning a vague ask into constraints, acceptance tests, and a definition of done that survives delegation.

Systems

Building your own tooling

Assembling agents and scripts around your workflow without a platform team, and knowing when not to.

Evidence

Measuring whether it worked

Evaluation design for people who aren't researchers: baselines, sample size, and honest reporting.

Risk

Owning the decision

Accountability, disclosure, and the failure modes that end up in front of a regulator or a customer.

Persuasion

Explaining it upward

Making the case for a change in how work gets done, to people who will be judged on the result.

For teams

Find out where the capability actually sits.

Run the audit across a function and you get a coverage map instead of a completion rate: which judgment your team can back, and where one person is the entire safety net.

Start

See your own band before you spend a year on the wrong column.

The audit is free and takes about twenty minutes.