Kirill Shvakov Autonomous Agents: transforming engineering and business

Autonomous Agents: transforming engineering and business

In the style of The Phoenix Project: from chaos to effectiveness with AI agents.

Status: Work in Progress. The structure is stable; chapters are being edited and refined.


How to read this book#

Fast start for engineers:

  1. Start with Chapter 1 — create your first working prompt and verification plan
  2. Use the “Quick start” section in each chapter for immediate practical value
  3. Come back to the full chapter when you need depth

For managers and business stakeholders:


Structure: 3 cases, 10 chapters#

Case 1: deployment chaos (Chapters 1–4)#

Case 2: the Brent bottleneck (Chapters 5–7)#

Case 3: the payroll incident (Chapters 8–10)#

Chapters

Chapter 1: first prompt + verification plan

How to write your first prompt as a contract and verify an agent’s output with a verification plan instead of taking it on faith.

Chapter

Chapter 2: system prompt + guardrails + dialogue SOP

How to turn a one-off prompt into a role with guardrails: constraints, stop conditions, and a repeatable dialogue SOP.

Chapter

Chapter 3: spec v1 + plan v1

How to turn a task into a manageable project: spec v1, a work plan, a risk register, and verifiable done criteria.

Chapter

Chapter 4: Phoenix Project architecture v1

Architecture design v1: components, contracts, tradeoffs, and verifiable decisions instead of arrow slides.

Chapter

Chapter 5: an agent-driven SOP for "design → PR"

A repeatable development process with an agent: from design and guardrails to PR, verification gates, and artifacts that reduce Bus factor.

Chapter

Chapter 6: operations and incidents

How to make incident response repeatable: SLI/SLO, triage, response scenarios, escalation criteria, and verifiable runbooks.

Chapter

Chapter 7: security and infrastructure

Security-by-design for agent workflows: threat model, guardrails, change plan, least privilege, and fast rollback.

Chapter

Chapter 8: eval dataset + golden tests

How to measure agent quality: eval datasets, golden tests, regressions, and a continuous improvement loop without "trusting the answer".

Chapter

Chapter 9: agent teams + governance

How to make multi-agent practice manageable: delegation, roles and checks, Skill Router, Git-first Agent Skills, and failure patterns.

Chapter

Chapter 10: capstone (full cycle)

A full end-to-end cycle—from business request to production operations—on one continuous case.

Chapter