Skip to content

Blog

Your coding agent is a free audit of your documentation

Code generation is solved, so the constraint is now context. Every wrong answer an agent gives points at a standard the team knows and never wrote down.

Written by

Context engineering

Published: 2 July 2026 Updated: 2 min read

An agent wrote Java 11 syntax into a Java 21 project. The team looked at it and someone said the sentence that mattered: “That is exactly what a new developer would do if we never documented our standards.”

It was not a one-off. The same session produced deprecated database queries. It wrote code that skipped the team’s testing practice. Every failure traced back to the same root. The information the agent needed was not written down anywhere it could read.

The constraint moved from code to context

Code generation is solved. The model writes fluent, working code on the first pass. So the constraint is no longer the code. It is the context: the standards, the conventions, the decisions that live in senior developers’ heads and nowhere else.

The agent shows every gap at once

Here is the useful part. The agent makes that invisible gap visible, on demand, for free. A new hire takes weeks to surface the same gaps, and surfaces them quietly, in a pull request someone half-reviews on a Friday. The agent surfaces all of them at once, in a session, where the whole team is watching. Every wrong answer is a precise pointer at something the team knows and never wrote down.

Read the failure as an audit result

That reframes the failure. The instinct is to treat the bad output as the agent being unreliable. It is more honest to treat it as an audit result. The agent behaved exactly like a competent new hire dropped into the codebase with no onboarding. It guessed.

The fix is not a better prompt. The fix is writing the standard down, in the codebase, where both the agent and the next human can find it.

Every gap you close helps the next hire too

And that is the compounding part. Every gap you close for the agent you also close for the next person who joins. The context that makes the agent stop guessing is the same context that makes onboarding faster. You were going to need it either way. The agent just sent you the invoice early.

What would your agent get wrong today, and what would that tell you about your onboarding?

Further reading

  1. AI tool fatigue is a missing context layer Teams count their AI tools when the list that decides is the second one: skills, commands, CLAUDE.md entries and context files. Where that list is rich, the tool list shrinks on its own. 1 min readShort post
  2. Context engineering is the standardization you could never sell Preparing context for AI agents forces the documentation and standardization architecture teams have pushed for years. The agent does not need it more than the organization did. It removes the resistance. 1 min readShort post
  3. One AGENTS.md for four AI tools is harness debt A single context file behaves differently in every harness and every model. The four kinds of harness debt we keep finding in teams, and what a Dark Factory does instead. 2 min readShort post

All 16 posts on Context engineering

ClientsCase study: A large DIY retailer in Germany

Test the constraint in your system

The Factory Readiness Assessment connects this research to your value stream, your data, and the next decision.

Training on this topic: Agentic Engineering for Developers and QA
See the assessment

Want to check one team first? The self-check: twelve yes or no questions, three minutes, instant result, no account. Take the self-check