Note · July 2026

The standard is
never on the slide.

On the thing AI erodes first — and why your roadmap doesn't measure it. A two-minute scroll.

Every business that runs on expertisehas a standard its people hold —without being told to.

It rarely gets written down. It's the reason a client trusts the work. And it's the first thing AI will quietly wear away.

Put a capable model next to agood expert, and every use of itis a real improvement.

The analyst gets anomalies flagged. The engineer gets a draft that's 90% right. The reviewer gets a summary faster than reading the whole thing. All genuinely better.

AI didn't lower the standard.It moved the standard —to a place nobody was watching.

The day the analyst stops noticing the thing that isn't on the benchmark, nothing breaks. The output still looks right. It ships.

The dashboard is green.The standard is gone.

A pilot measures faster, cheaper, more accurate on a benchmark. It never measures whether your expert still notices the thing the benchmark can't see.

You can't put judgement on a slide.So it gets no column in the plan.

And what has no column in the plan is what a plan quietly assumes will take care of itself. It won't. By the time the erosion is visible, it's already in the product.

The better question

Not “where can we deploy AI?”but: where would it let a goodperson stop doing the thing

we trust them for — andhow would we even know?

The argument, in four lines

AI commoditises the middle. Value migrates to the extremes.

AI moves the bottleneck from creation to verification.

AI scales creation. It doesn't scale judgement.

Will “human-made” command a premium in your industry — the way “organic” does in food?

What to actually do

Three moves — none of which slow you down.

Find where the judgement lives before you automate near it.

Sit with the people trusted for it — while they work, not in a workshop about it. The standard is almost never where the org chart says it is. It's usually one person doing one thing everyone else has quietly come to rely on.

Keep the human on the part that carries the standard, not the part that's tedious.

Automate the tedium. The danger is only when the tedious act and the standard are the same act — when reading the whole thing is both the boring part and the reason nothing bad ships. Leave that where a person still exercises it, and knows they do.

Make the name on the work stay human.

When someone's name is on the output — when they'll be asked to stand behind it — the standard holds, because someone is accountable for it. Verification isn't a step bolted on at the end. It's what keeps the standard from drifting somewhere nobody's watching.

AI should strengthen the expertise your business runs on, not quietly stand in for it. The difference is rarely visible in a demo. It's visible in the work — which is the only place worth looking.

One thing you can do this week

Sit with one of your best people for an hour while they do the thing they're trusted for. Write down every judgement call they make that isn't in the process.

What it won't tell you

You'll have one person's version of the standard. You won't yet know which judgements are shared, which differ, or which differences actually matter.

If this is live for you.

Send me the thing you can't get a straight answer on. I'll tell you whether I'm useful.

Email me →