10. Practitioner Video Evidence
This ledger connects the architecture to dated practitioner and product-creator testimony. Every quoted sentence below was checked against the local WebVTT transcript and links to the corresponding YouTube timestamp. Capitalization and punctuation are normalized; wording is preserved. The transcript proves what was said at that point in the video. It does not prove that a product still behaves that way, that a reported outcome generalizes, or that an interview claim was independently measured.
Use three evidence levels throughout this page:
- Official current evidence for product behavior, limits, licences, and availability.
- Creator interviews for dated design rationale and historical product decisions.
- Practitioner talks for methods, failure reports, and hypotheses that still require local validation.
| Source | Timestamped verbatim | What it supports | Evidence boundary |
|---|---|---|---|
| Harness Engineering: The New Discipline of Agentic Dev, AI Native DevCon interview, 2026-07-08 | "Never tell the agent more than once how to do a thing before it gets codified somewhere in the inner or outer loop." | Inner, outer, and meta feedback loops | Practitioner model, not a formal standard; speaker name is not established by the local metadata |
| Patrick Debois, The Rise of Agent Enablement, 2026-07-08 | "We're not building the thing, we're building the thing that builds the thing." | Agent enablement as system design | Organizational framing, not a benchmark result |
| Ryan Lopopolo, Harness Engineering: How to Build Software When Humans Steer and Agents Execute, 2026-06-19 | "I never want to give the same review feedback twice." | Codifying recurring feedback into the harness | Practitioner workflow and definition |
| Marc Sloan, Harness Engineering Beyond Code, 2026-06-20 | "Product and design context is important to agents, but it lives outside the code base." | External product and design context | Practitioner talk; integration trade-offs require local testing |
| Boris Cherny, Building Claude Code, 2026-03-04 | "If you're not sure, just ask the human and then they can decide." | Conservative permission fallback | Creator interview; current behavior belongs in official docs and release notes |
| Dax Raad, Building OpenCode, 2026-05-27 | "You need some kind of control plane to like set up all the providers, permissions, budget controls, rate limits." | Separation between runtime and organizational control plane | Creator interview and dated product claim, not proof of current availability |
| Simon Willison, Engineering Practices That Make Coding Agents Work, 2026-03-19 | "I think tests are no longer even remotely optional." | Tests, proof artifacts, and sandboxing | Practitioner recommendation |
| Shachar Azriel, Executable Specs: Building a Verification Layer for Agentic Coding, 2026-06-15 | "The code is how we ground that agent to reality." | Requirement-level verification against actual code | Architecture described by the speaker; outcomes are not independently verified here |
| Amit Kushwaha, Benchmarking the Agent Era, 2026-07-10 | "You can't talk about mean and medians. Now you have to talk about in terms of distributions." | Multi-turn, multi-tool, tail-aware measurement | Benchmark-methodology talk; specific performance claims require the underlying study |
| Dex Horthy, Context Engineering, 2026-07-15 | "If you want to do loops engineering, you should build one loop at a time and keep them small and contained." | Bounded loops before lights-off automation | Practitioner interview and retrospective, not a controlled study |
| Pavan Belagatti, What Is a Software Factory?, 2026 | "Agents do the work, humans provide the gates." | Separation between the execution loop and the governance loop | Practitioner short and Port-oriented framing; no comparative reliability or productivity measurement |
The source list is selective. It includes videos that add a distinct mechanism, metric, or failure boundary to this page. Inclusion is not an endorsement of every claim in a talk.