Home/Knowledge Base/Knowledge Worker Playbooks/K-04

K-04K-Series · Foundation

The Traffic Light System for AI at Work

Not every AI-assisted task carries the same risk, so not every task deserves the same caution. The Traffic Light System gives you three levels — green, amber and red — based on two things: how badly an error would hurt, and how easily you’d spot it. It’s the fastest way to right-size your oversight, and the fastest way to explain your reasoning to a manager, a team or an auditor.

Why this matters

Blanket rules fail in both directions. ‘Never use AI’ hands back the hours your competitors are saving on routine drafting. ‘Use it for everything’ eventually publishes an invented figure in something that matters. What organisations and individuals actually need is proportionate oversight — heavy where errors are consequential or hard to detect, light where they’re trivial and obvious. Three colours encode that in a form everyone remembers.

The three levels

• Green — safe for AI with verification. Errors would be easy to spot and cheap to fix. Routine emails, internal summaries of documents you know well, first drafts you’ll rework anyway. Your check: a careful read before it leaves you.
• Amber — use AI carefully, human supervision required. Errors are harder to spot or the audience matters. Executive briefings, client-facing documents, anything with figures you didn’t produce yourself. Your check: verify against sources, and a second set of eyes where stakes justify it.
• Red — human judgement required. Errors would be costly, hard to detect, or the task is judgement itself: decisions about people, legal positions, confidential material without an approved tool. AI’s role here is preparation at most — never the work itself.

The two questions behind the colours

When you’re unsure of a colour, ask: how bad is an undetected error here, and how likely am I to detect one? Consequence and detectability. A typo in a team lunch invite is trivial and obvious — deep green. A plausible-but-wrong number in a board paper is consequential and nearly invisible — firmly red territory for anything unverified. Most tasks sort themselves the moment you ask both questions.

One caution from long experience with this system: tasks drift. The weekly report that was green internally becomes amber the day it starts going to the executive. Re-colour when the audience or stakes change, not just when the task does.

Putting it into practice

1. Colour your five most frequent tasks right now — write the colour next to each
2. For each amber, define what ‘supervision’ actually means before the next occurrence
3. For each red, decide what preparation help (if any) AI may give
4. Agree colours with your team so oversight is consistent, not personal
5. Review colours when a task’s audience changes — that’s when drift happens