harness-digest

A weekly read,
picked by a harness.

Twice a week a small agent harness reads the web for writing on agentic engineering. Sonnet finds and summarises. A separate Opus judge decides whether each piece has a real teaching moment. Plain code checks every quote against the source and enforces the quotas, so the prompt doesn’t have to. On Sundays the best five to seven become an issue.

  1. 7 picks from 30 that passed the judge

    Week 46

    Every item this week deals with the same gap: agents produce changes faster than people can verify them, and the fix is to move verification out of human attention and into the structure around the agent.

  2. 7 picks from 21 that passed the judge

    Week 45

    Most of this week's items argue that agent reliability comes from the scaffolding around the model rather than the model itself, and that the scaffolding only counts if it is enforced.

  3. 7 picks from 28 that passed the judge

    Week 44

    Most of this week's items replace an agent's own say-so with something that can be checked: screenshots captured automatically, evidence required in every report, stop hooks, a fresh-context reviewer, and guardrails enforced in code at the tool-call layer rather than in the prompt.

  4. 7 picks from 20 that passed the judge

    Week 43

    Most of this week's items argue that agents get better when you give them less to do and a clearer way to check their work.

Each issue is also committed to the repo as markdown. Read them there.