Thinking in Agents and AI-Native Systems: Designing, Building and Governing Agents on a Deterministic Core

An agent is a thing in code. Its autonomy level is the shape of that code. Its safety lives somewhere the model can never reach.

Thinking in Agents and AI-Native Systems is the engineer's book of the series: build four agents at autonomy levels 1 to 4 from scratch, run an offline simulated model so every listing runs on any laptop with no API key, write handoff contracts and a coordinator with step limits, put the never list into a deterministic core that refuses whoever proposes — agent or person — with no bypass, give every agent one identity and exactly its grants in one shared store, record every decision so it can be replayed for a regulator, and assemble a native process from five parts with a stop rule computed from the chain's own numbers.

Every listing is embedded from a tested code package (about 1,200 lines, nine tests, a reference architecture script) and every output printed in the book was produced at build time. The reference architecture is Cedar Bank's SME credit process: $310 to $95 a file, four days to four hours, 2.5x volume on the same forty analysts.

After this book you will be able to:

• Write an agent specification as data and register agents that a runtime refuses without an owner
• Build agents at every level and see the level as the shape of the code
• Chain agents with handoff contracts, a coordinator, and controls for loops, conflicts and silent errors
• Build a shared store with per-agent identities, exact grants and a write trail
• Encode a never list as rule functions with owners, versioned and changed only by signature
• Implement checkpoints, decision logs, replay, three numbers and per-agent indicators
• Assemble a native process, migrate by cohort, and hand the board the numbers it asked for

26 chapters, 26 figures, 208 self-test questions with answer keys, the full code in Appendix A, templates and an 18-item architecture checklist in Appendix B. Book 5 of the Thinking in AI series.

In this series: 1 The Foundations · 2 Thinking in AI for Leaders · 3 Thinking in AI for Managers · 4 Thinking in AI at Work · 5 Thinking in Agents and AI-Native Systems · 6 The AI-Native Casebook.