AAI Agentic Handbook
Methodology

How this handbook is written

What you can rely on, what changes fast, and how mistakes get fixed.

Who it's for

Everyone building with AI: people with no hands-on experience, new engineers, and senior and staff engineers preparing for design reviews or interviews. Every page carries a level (Curious, Budding, Mid, Senior, Staff) so you can tell who it was written for.

How content is produced

Drafts are written with AI assistance and then edited against a written quality bar: commit to a default and say when to change it; state numbers with the conditions that make them true; name real tools and models; include a realistic failure and what to monitor; cite a primary source. The full bar is published with the source code.

Byline: content is published as AI Agentic Handbook editorial. We don't invent personal experience. Worked STAR stories are composite examples written to show structure, and they are labelled as illustrative.

Sources

Term pages, interview questions and posts list their primary sources: papers, specifications and official documentation. Where a claim depends on a vendor's current pricing or limits, it is stated as an assumption with a date.

Freshness

Every page shows when it was last reviewed. Pages are tagged by how fast their facts change:

  • Stable — mathematics and algorithms (attention, BM25, HNSW). Reviewed yearly.
  • Yearly — architectures and practices (RAG patterns, eval methods).
  • Quarterly — prices, model limits, vendor features and protocol versions (marked “fast-moving”).

Changes are recorded in the changelog.

Corrections

Found something wrong or out of date? Use the “Was this helpful?” control at the bottom of the page and choose “Not really” — those signals are reviewed with every content update. Substantive corrections are listed in the changelog with what changed.

What it isn't

Not vendor documentation and not sponsored. Product names appear because engineers need to know what to reach for, not because anyone paid for a mention. Company tags on interview material describe a format, not a leaked question.

Privacy

Your progress, quiz history and level choice are stored only in your browser. Page analytics are cookieless and aggregate.