AE

AI Engineer Press

Ideas for progress

From Copilot to Colleague cover
Isachenko · MohanraoFrom Copilot to Colleague
AE

Tap the spine to open the book

◇

Immersive journey

Fly through all 10 chapters in 3D

Definition

What is "From Copilot to Colleague"?
"From Copilot to Colleague" is a source-anchored book on AI engineering: how AI systems move from assistants that suggest, to copilots that collaborate in real time, to delegates trusted with full, reviewable tasks. Every claim links to a timestamped moment in one of 941 AI Engineer conference talks — 54 claims backed by 199 source anchors — and every chapter is scored by a panel of three rival open models (the book's own measurement engine, book-mash). Written by Timur Isachenko & Daniel Mohanrao, published as an evolving public experiment at fromcopilottocolleague.com.
What makes this book's method different from other AI commentary?
Most AI commentary is assertion without provenance. This book's claims layer works the other way: every claim in the ledger carries a Source Anchor — a precise video id and timestamp pointing at the moment a practitioner said it — plus a support level (strong, moderate, or tentative) instead of a flat citation. The claims ledger, chapter drafts, and chapter-by-chapter judge scores are all public and machine-readable, so the method is reproducible, not just the manuscript it produced.

claims-ledger · open source

CI that fails when docs lie

Same claim grammar as this book — now for your codebase. Every strong claim carries a verbatim quote anchor; stale pointers exit 11 in CI.

Claims verified badge
WebsiteGitHub repoFact-checked ledgersFork sandbox

About the Lab

I've been running an experiment.

The Lab is a public project to produce a real, source-backed book on AI engineering by running autonomous research-and-writing loops over a corpus of practitioner talks. The AI Engineer YouTube channel — currently 794 videos and growing — is the primary source brain. Live counts in [STATS.md](https://github.com/isatimur/ai-engineering-book-lab/blob/main/STATS.md).

The deliverable is not just The Manuscript (the chapter drafts you're reading). The deliverable is The Method — a reproducible research-and-writing machine that turns source material into structured notes, theme and concept synthesis, verified Claims with Source Anchors (precise pointers: video id + timestamp), chapter packets, and — eventually — a finished book.

The field evolves too fast, the noise is too high, and the same ideas get repeated in slightly different packaging every week. is partly about writing a book and partly about building a system that helps me stay current, separate signal from noise, and surface the patterns that already have evidence behind them.

The workflow today:

videos → → → chapter drafts → public iterations

And the direction I'm pushing it in, autoresearch-style:

  • bounded research passes
  • self-improving instructions
  • source-fidelity checks
  • quality judges for summaries, , and chapters
  • coverage, coherence, and drift metrics

The Manuscript is the visible output of a larger experiment: can a book become a public, self-improving research artefact?

This started as "let's make better notes." It's turning into something closer to a public experiment in writing, judging, and continuously improving that produces — with the discipline that no claim ships without a source anchor.

For AI agents

This book is structured for LLMs and agents to read — a machine-readable index and the full text as clean markdown.

llms.txtllms-full.txtsitemap.xmlEvent ledgersclaims-ledger

Ask AI about this book

Open any AI assistant with a pre-filled prompt about the book, its method, and what makes it different.

✦ Claude✦ ChatGPT✦ Perplexity
Volume I