Agentic Coding Teams

What the Evidence Says About Teams of Humans Working with AI Agents

The conversation about AI in software development has collapsed into two unhelpful extremes: the claim that nothing important has changed, and the promise that human teams are about to disappear into a fully automated "dark factory". The truth lies in the middle: in the real teams that already use agents as collaborators every day. This book is a calm, evidence-based guide to how those teams actually work, and to what the current research genuinely supports, rather than what the marketing on either side would have you believe.

Format: PDF / 92 Pages
Language: English
Authors: Pierluigi Pugliese and Dario Zanotto

The Book:

Every substantive claim in these pages has been checked against its source, not merely that the source exists, but that it says what is claimed. The evidence is deliberately weighted towards primary research: peer-reviewed and preprint studies, large-scale developer surveys, and industry programmes that publish their methodology, rather than vendor marketing or consultants' forecasts. Where a widely repeated figure did not hold up, it was corrected or set aside.

What emerges is a practical picture. Small teams of four to six people remain the sweet spot, and they are the heaviest and most effective adopters of these tools, not the laggards. The hard work of software does not disappear when an agent writes the first draft; it moves from writing code to reviewing and verifying it, and that burden lands on your most experienced people. More output is not more value if the review bottleneck cannot keep up.

The book spans the team and its practices, the hidden costs of working with agents (cognitive load, cognitive fatigue, the real cost of tokens, and the security tax), the changing role of the Scrum Master and Agile Coach, and the governance and sovereignty questions that AI raises, from regulation and intellectual property to the case for keeping AI inside your own infrastructure. It closes with a measured answer to the dark-factory claim and a note on how every fact was verified.

None of the sensible responses depends on predicting the future. They are the disciplines good organisations already valued: clear standards, a shared language, protected review, defended collaboration, now made non-optional, because an agent amplifies whichever you choose: your standards or their absence.

A note on how this book was made:

This book was written with heavy use of AI, which is deliberate and fitting for its subject. The content is nonetheless checked with care: every substantive claim has been verified against its source, and the whole has been read, corrected, and validated by the authors against their own experience of working with these tools. The AI produced much of the draft; the judgement about what is true and what is worth saying remained the authors'.