How to Build Autonomous Coding Agents

Reading paths

Shorter routes through the book. Each one stands on its own.

Contents

9 sections in reading order. Each row gives the part's claim and how long it takes.

1. I

   Before you start

   How the parts fit and where to begin.

   3 min in 1 part

   1. Start hereHow to use this bookRead the two parts that match your problem this week. Come back for the rest.3 min

2. II

   The economics

   Most of the token bill comes from the system around the model.

   5 min in 1 part

   1. 1AI agents are not expensive. Bad architecture is.Most of an agent's token bill comes from the system around the model, not from the model.5 min

3. III

   Deterministic retrieval

   Decide in code what the agent reads.

   16 min in 4 parts

   1. 2Stop making agents read your entire logsLogs are structured data, and treating them as prompt text costs tokens you do not have to spend.4 min
   2. 3Parse the user prompt before the model ever sees itYour prompt already carries structured information, so extract it before you pay for inference.4 min
   3. 4Why grep is the wrong retrieval engineSearching is not understanding, and code already carries the structure you need.4 min
   4. 5Build semantic memory onceA run should start from what the last one learned instead of repeating it.4 min

4. IV

   Context engineering

   Decide in code what stays in the window.

   13 min in 3 parts

   1. 6A rule written in prose cannot fail CIConstraints belong in a representation a validator can check.4 min
   2. 7Context compression is worth more than a bigger modelRemoving irrelevant information beats buying a larger model.4 min
   3. 8Agents guess at schema they were not shownA model adds a duplicate column when the context did not carry the existing one.5 min

5. V

   Orchestration and memory

   Decide in code which step runs next and what the agent remembers.

   18 min in 4 parts

   1. 9Why coordinator agents don't scaleMost multi-agent systems pay a coordination tax and call it architecture.5 min
   2. 10Tests should be first-class retrieval objectsTests describe behavior in executable form, and a test runner's verdict is computed rather than inferred.4 min
   3. 11Agents need working memory, not bigger context windowsYou do not reread every book you own before fixing a bug, and your agent should not either.4 min
   4. 12Knowledge graphs beat prompt engineeringKnowledge should be connected, not concatenated.5 min

6. VI

   Measurement

   Measure the agent the way you measure a service.

   5 min in 1 part

   1. 13Measuring agent intelligenceBenchmarking on coding challenges misses what matters in production.5 min

7. VII

   Operating the workforce

   Identity, authority, budget, equipment, and a record another person can read.

   33 min in 7 parts

   1. 14Give every agent its own identityYou cannot set authority for, bill, or review something you cannot name.4 min
   2. 15Write the mandateFour decisions govern an agent. Most teams have made all four. Few have them in one place.4 min
   3. 16The agent asks, a rule decidesAuthority is a decision at the moment of use, written down by the team that owns the system.5 min
   4. 17Spend you can attributeA total is not an answer. The answer is which agent spent what, and on whose behalf.5 min
   5. 18Equip agents on purposeA tool in the catalog is not a tool in the agent's hands. Assign equipment the way you assign access.5 min
   6. 19Keep a record another person can readA log answers the engineer who wrote it. A record answers the person who was not there.5 min
   7. 20Bounded tasks and ongoing workSome work has an endpoint. Other work continues. Manage each in its own way.5 min

8. VIII

   The endgame

   Where the system goes next, and a closing thought.

   5 min in 2 parts

   1. 21Better systems, not better modelsYou rent the model. You own the system around it, and the way you operate it.4 min
   2. ClosingClosing thoughtIntelligence is expensive. Determinism is cheap. Spend the first only where the second cannot do the job.1 min

9. IX

   Back matter

   Templates, terms, and sources.

   1 min in 1 part

   1. Field kitField kitTemplates and worksheets to copy. Print this section and fill it in with the people who own each line.reference
   2. GlossaryGlossaryThe words this book uses in a fixed sense.reference
   3. About the authorAbout the authorOxagen is workforce management for autonomous agents: give each agent an identity, set its authority and budget, equip it with tools and skills, and review…1 min
   4. SourcesSourcesEvery empirical claim in this book traces to one of the sources below. Read the primary literature. It is better than any summary of it, including this one.…reference