← All ledgersEvent ledgerBook graph

Sample ledger — honestly labeled; sourced from the book corpus, not live-extracted at publish time.

Event ledger — Ryan Lopopolo, OpenAI

Harness Engineering: How to Build Software When Humans Steer, Agents Execute

5 claims · 5 timestamp anchors · 1 speaker. Watch the full talk: YouTube.

Claims & source anchors

The important thing is not the code but the prompt and the guardrails that got you there

strong
  • The important thing is not the code but the prompt and the guardrails that got you there.
    Ryan Lopopolo, OpenAIconfidence: high

Documentation, ADRs, and breadcrumbs teach agents what good work looks like

strong
  • leaving breadcrumbs, documentation, ADRs, persona oriented documentation around what a good job looks like.
    Ryan Lopopolo, OpenAIconfidence: high

A good harness operationalizes giving the model text at the right time

strong
  • a good harness is really operationalized around giving the model text at the right time
    Ryan Lopopolo, OpenAIconfidence: high

The practical unit of harness engineering is the codebase, not isolated snippets

strong
  • codebase for harness engineering
    Ryan Lopopolo, OpenAIconfidence: high

Reliability comes less from model cleverness than from surrounding scaffolding

Book synthesis claim — anchored to this talk as primary evidence.

moderate
  • The important thing is not the code but the prompt and the guardrails that got you there.
    Ryan Lopopolo, OpenAIconfidence: high

Powered by claims-ledger — machine-verifiable pointers with CI-gated verification. No anchor, no strong claim.