AI Engineering Blogs: Five Sources for Systems, Evals and Practical Work
ZarifAn engineering reading list should help with a decision: what to measure, what to build, or what to change when a system fails. These five writers serve different parts of that loop. This is a deliberately small selection, not a traffic ranking.
Last verified September 17, 2026. Rechecked monthly. Download CSV (5 entries).
Start with five
Use Simon Willison for a quick experiment, Hamel Husain for evaluation practice, Chip Huyen for system design, Eugene Yan for applied methods, and Lilian Weng for research foundations. Follow the primary references in an article before treating a result as transferable to your workload.
- Simon Willison · Frequent short posts; skim weekly
Read for inspectable experiments: prompts, outputs, small programs and links back to the release being tested. Useful when deciding what to try locally.
- Hamel Husain · Irregular long-form posts; check monthly
Start here when an AI feature works in a demo but fails for users. The writing emphasizes error analysis, realistic evaluations and the decisions those measurements support.
- Chip Huyen · Irregular essays; check monthly
Read for system design across data, evaluation and deployment. Especially useful before choosing infrastructure around a model.
- Eugene Yan · Irregular technical articles; check monthly
Read for the connection between an evaluation method and a product outcome, with detailed examples from retrieval, recommendations and LLM applications.
- Lil’Log — Lilian Weng · Occasional deep dives; keep in RSS
Use the cited papers and technical explanations to understand a method before adopting its fashionable label. Best for a slower study session.
The directory
Selection favors inspectable methods, named tradeoffs and links to supporting work. Publication schedules are irregular; the cadence below describes a suggested reading routine, not a promise from the author. Pages and public feeds were checked on the verification date.
| Name | Role | Why it is here | Cadence |
|---|---|---|---|
| Simon Willison | Tools and implementation | Read for inspectable experiments: prompts, outputs, small programs and links back to the release being tested. Useful when deciding what to try locally. | Frequent short posts; skim weekly |
| Hamel Husain | Evaluation and product engineering | Start here when an AI feature works in a demo but fails for users. The writing emphasizes error analysis, realistic evaluations and the decisions those measurements support. | Irregular long-form posts; check monthly |
| Chip Huyen | Production AI systems | Read for system design across data, evaluation and deployment. Especially useful before choosing infrastructure around a model. | Irregular essays; check monthly |
| Eugene Yan | Applied ML and evaluation | Read for the connection between an evaluation method and a product outcome, with detailed examples from retrieval, recommendations and LLM applications. | Irregular technical articles; check monthly |
| Lil’Log — Lilian Weng | Research foundations | Use the cited papers and technical explanations to understand a method before adopting its fashionable label. Best for a slower study session. | Occasional deep dives; keep in RSS |
Turn one article into an experiment
Pick a current problem before opening your reader. For an unreliable assistant, that might be whether retrieval missed the evidence or the model ignored it. Read one relevant article, record the proposed mechanism, and test it against a small set of real failures. Save the result beside the article link, including where the advice did not fit.
That habit is more useful than subscribing to every launch feed. Revisit the list monthly and remove sources that no longer change a decision.
The RSS and OPML pack offers five working subscriptions, with alternatives where a publisher’s feed is stale or has unusable item links. For lab announcements and reported coverage, use the broader AI blogs and news directory. To move from reading to code, choose one agent repository.
Change log
Added a focused selection with primary-source links, a start-with-five and an export.