See It Work — Ninety percent of agent failures are context failures | Book 02 · AI Agents for Executive Decisions

When a decision agent disappoints, the instinct is to blame the model. Usually wrong: ninety percent of agent failures are context failures, not model failures. The same model produces radically different output depending on what you feed it — so the real skill, context engineering, is getting the inputs right.

Ninety percent of agent failures are context failures — full detailed chart

The full detailed chart. Condensed for print legibility in the book; shown here at full size.

Chase a 'better model' and you fix the wrong thing. Fix the context — what the agent is given — and the same model suddenly performs, because the inputs were the problem all along.

What this means for you

Most agent failures are context failures — fix the inputs, not the model. What this means for you: you stop wasting money chasing fancier models to fix bad results, and start fixing the actual cause — the context you feed the agent — which makes the model you already have perform, because the inputs were the problem.

The failure is in the inputs, not the model:

Context Engineering

blamedthe model

actuallythe context

same modeldifferent context → different output

the skillget the inputs right

Ninety percent of agent failures are context failures, not model failures.

For the technical reader — the command, and how to verify it yourself

# one line · you do not need to run this
see walkthrough

see walkthrough
# -> the same model performing once the context is engineered right

Full step-by-step is in Appendix RX: Hands-On Demonstrations in the book.