FIELD BRIEF · AI OPERATIONS

The step your SOP forgot to write down.

Most operating docs are written once, by someone who already knows the context — then handed as-is to an AI agent that doesn't. The fix isn't a longer document. It's writing the why, flagging the judgment calls, and making it as easy to update as it was to write.

BY CAROLINE VRAUWDEUNT · CEO, ANDRS · FULL ARTICLE · SEP 2026

WHAT IS A FIELD BRIEF?

A Field Brief is a structured, one-page distillation of a key strategic question. Each brief diagnoses the root cause, maps where the gap shows up, and closes with a concrete next step — designed to be read in under 5 minutes and acted on immediately.

Risograph-style illustration of a spool of gold thread mounted by a door, unspooling through the doorway and trailing across the floor

THE DIAGNOSIS

is how many of its own blind spots an AI will flag on its own — it writes instructions that sound complete but quietly assume knowledge only you have, unless you explicitly ask it to check.

FIG. 05 · THE CONFIDENCE PROBLEM

A rigid SOP handed to an AI agent doesn't fail loudly. It does the wrong thing confidently — the same way a rigid guideline handed to a new hire used to.

THE / THESIS

"Most SOPs are written for the what — the steps. The one that survives contact with an AI agent is written for the why — the reasoning behind them."

Where a rigid SOP breaks — and the four-part fix that keeps it flexible.

01 — RIGID VS. FLEXIBLE

01 · The why

Write the reasoning, not just the steps.

A rigid list of steps tells an AI agent what to do. It doesn't tell it what to do when the situation doesn't quite match — which is most situations. A reason gives whoever follows it room to adapt.

TELL: THE DOC ANSWERS WHAT, BUT NEVER WHY

02 · Judgment

Flag the judgment calls, out loud.

Not every step should be followed blindly. Mark the 1-2 steps where a human should decide — explicitly — so neither a new hire nor an AI agent executes them on autopilot.

TELL: NOBODY CAN POINT TO WHICH STEP NEEDS A HUMAN

03 · Updating

Make updating as easy as writing.

Use AI in reverse — dictate the change, ask it to update the doc — instead of letting the process drift out of date the way guidelines quietly always do.

TELL: THE LAST EDIT WAS MONTHS AGO AND NOBODY NOTICED

04 · Timing

Review on a trigger, not a calendar.

Update it when someone new joins or a process breaks — not on a schedule nobody actually keeps. A calendar reminder gets snoozed. A broken process gets noticed.

TELL: THE REVIEW DATE KEEPS GETTING PUSHED

The Move — write one down.

02 — ONE TASK, ONE PROMPT

PRESCRIPTION · 10 MINUTES

Pick one task you keep re-explaining. Talk it into your phone, then hand Claude the prompt.

R.01

If the output reads complete: check whether you explicitly asked it to flag vague steps and judgment calls. If you didn't ask, it didn't check — that's on you, not the tool.

R.02

Once it's written: save it somewhere your AI remembers it — a Skill, a Custom GPT, a Gem — so you stop re-explaining it every time the task comes up.

→ BONUS: DICTATE CHANGES BACK INTO IT INSTEAD OF LETTING IT GO STALE

From the field.

03 — SEVEN YEARS LATER

A FOUNDER CASE STUDY

IN-HOUSE

Why the 2019 fix finally works — now that AI can read it.

In 2019 I wrote four tips for team continuity — tools, guidelines, logging, open culture. Good advice, hard to keep living up to. AI didn't replace the advice. It made actually following it easier.

LEVER 01

Skills over static docs

We keep our own know-how in Claude Skills instead of a wiki nobody opens — so it stays active in the work, not filed away.

LEVER 02

Shared memory, not one head

We're investing in a shared AI memory system so the why behind a decision doesn't live in one person's head alone.

THE BOTTOM LINE

The fix was never a longer document. It's one your team — human or AI — can actually keep alive.

NEXT STEP · FREE

Want to see if your team's know-how is AI-ready?

consultancy.andrs.nu→