Lesson 2 of 6 · 10 min
What it's good and bad at today
After this lesson: Tell a 'green' task from a 'red' one before you try it.
AI is not uniformly clever. It's excellent at some things, useful-with-care at others, and genuinely unreliable at a few. We sort tasks with a simple traffic light — the same one we use across the site.
Green — go
Reliable enough to use now, with normal care. Drafting and rewriting text, summarising long documents, brainstorming options, explaining a concept in plain English, tidying messy notes. A person still reads the result, but the heavy lifting is safe to hand over.
Amber — caution
Useful, but keep a human firmly in the loop and check the output before it ships. Anything with numbers, names, dates, quotes, or legal and regulatory wording. The tool will sound certain; that confidence is not evidence.
Red — not yet
Promising but not dependable. Don't bet the business on it today: unsupervised decisions that affect people, fully automated advice in regulated areas, anything where a wrong answer is expensive and you wouldn't catch it.
Browse the full AI Reliability Traffic LightOur running, dated register of what AI is ready for — and what it isn't.