The standards we work to, free to take and use.

Everything here is the actual working material from our engagements, not a summary of it. No form, no email, no attribution required. If it is useful without us, take it and use it without us.

Resources

Three artifacts, available now

These are the documents an engagement produces. We publish them because a standard that only works while we are there is not a standard. The shorter files — skills.md, gate.md — are also real URLs. An agent fetching this site should start at llms.txt.

Template

Approval page

Eight sections a job must answer before software may act on a real system — what may be drafted, who still sends, thresholds, record, how you undo it, kill switch and sign-off. The file is gate.md. The document your risk function signs.

Instrument

Baseline worksheet

The five variables to capture before anything changes — unit of work, touches, minutes, rework rate, cycle time — with the attribution rule that stops you claiming a seasonal dip as a result.

Checklist

Pre-production review

Twenty-two questions to run against any agent before it is permitted to write. Use it on ours, on your own, or on a vendor's.

The maturity model

Most companies we meet are at stage one or two — Assisted or Equipped. The five stages of how a company uses AI are on the home page. The jump is a named person who still sends, and a record of it.

See where most sit
Writing

Notes from the part nobody demos

Short pieces on what actually stops enterprise AI reaching real work. Published here and, where relevant, in the business press.

Date

Your pilot worked. Why can't you switch it on?

The six-month gap between a working prototype and a live deployment, and the five artifacts that close it.

In draft
Date

A gate that always approves is not a gate

Why approval rate is the wrong metric, what edit rate and reject rate actually tell you, and how ceremonial sign-off passes audit while protecting nobody.

In draft
Date

What "20% more efficiency" has to mean before you sign it

Four sources of efficiency, five variables, and the attribution rule that separates a real gain from a good quarter.

Planned
Date

Most AI training changes nothing. That is a design choice.

Four decisions that make the difference between a session people enjoyed and a session that changed how work gets done.

Planned
Date

Nobody jumps to stage four

Why the maturity ladder cannot be skipped, and what breaks in the organisations that try.

Planned

Before publishing

An insights section showing five "planned" items reads worse than none at all after about a month. Ship the first two pieces before this page goes live, or hide the section until they exist.

Next step

Ninety minutes. One piece of work. A written answer.

Take the templates and use them yourself. If you would rather have someone with you while you do, that is what the ninety minutes is for.

Book the ninety minutes
Book the ninety minutes