--- schema: 1 type: video title: >- The Last Human Code Review: Building Trust in AI-Generated Code — Itamar Friedman, Qodo description: >- If you are shipping AI generated code faster than your humans can review it, Itamar Friedman's position is that you are inside the problem rather than ahead of it. He asks the room whether developers will still be reading diffs line by line by the end of the year, then reports finding two incompatible camps among engineering leaders the night before: one holding that every line must be human trusted, the other content to ship bugs and fix them quickly because velocity wins. Which camp you sit in decides what you have to build. His claim is that models stopped being the constraint. Code review benchmarks have barely moved across recent model releases, and the difference between a review that catches a real contract break and one that asks whether you considered error handling is context, not reasoning. That context is scattered across competing instruction files, differs between teams inside the same company, and largely is not written down at all, living instead in senior developers' heads and in Slack threads. Codifying it means building for two audiences at once, because the format an agent parses cleanly is not the format developers will actually maintain. The deeper version encodes the architecture itself, including which service contract broke production three months ago, so that review shifts from reading one pull request to reading a graph and noticing that three changes in flight are about to collide. Speaker info: - https://twitter.com/itamar_mar - https://www.linkedin.com/in/itamarf - https://www.qodo.ai/authors/itamar-f Timestamps: 0:00 - Where the bottleneck moved, and why code review exists 3:38 - Two camps: trust every line, or ship and fix fast 5:19 - Models are not the barrier, context is 7:03 - Context scattered across competing instruction files 8:48 - The knowledge lives in heads and in Slack 9:42 - Codifying for agents and humans at once 10:34 - Interfaces for both: rules shown, and a note to the next agent 12:19 - Fewer human comments as the readiness signal 13:11 - Encoding architecture, contracts, and past outages 14:05 - Automatic approve and block, added gradually 16:38 - Reviewing the software graph instead of the PR --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/s-aixZYJG4c-the-last-human-code-review-building-trust-in-ai-generated-code Join our upcoming AI Engineer conferences (see https://ai.engineer landing page): - AIEi Shanghai, November 5–6 — the first for meeting top Chinese labs and engineers - AIE CODE in SF, November 10–12 — the top AI Coding conference returns - AIEi Sydney, December 7–8 — the first AIEi run alongside NeurIPS 2026 in Sydney --- canonical: >- https://liquidgold.cc/channel/aidotengineer/explore/video/the-last-human-code-review-building-trust-in-ai-gen-74c11726ce36 markdown: >- https://liquidgold.cc/channel/aidotengineer/video/the-last-human-code-review-building-trust-in-ai-gen-74c11726ce36.md llms: null parent: 'https://liquidgold.cc/channel/aidotengineer.md' channel: aidotengineer updated: '2026-10-24T05:07:34.362Z' language: null children: [] videoId: s-aixZYJG4c durationSeconds: 1134 publishedAt: '2026-08-20T13:30:38.000Z' searchChunks: - ordinal: 0 text: null startSeconds: 12 endSeconds: 218 watch: 'https://liquidgold.cc/watch/s-aixZYJG4c?t=12' heldReason: rights - ordinal: 1 text: null startSeconds: 12 endSeconds: 218 watch: 'https://liquidgold.cc/watch/s-aixZYJG4c?t=12' heldReason: rights - ordinal: 2 text: null startSeconds: 218 endSeconds: 229 watch: 'https://liquidgold.cc/watch/s-aixZYJG4c?t=218' heldReason: rights - ordinal: 3 text: null startSeconds: 329 endSeconds: 340 watch: 'https://liquidgold.cc/watch/s-aixZYJG4c?t=329' heldReason: rights - ordinal: 4 text: null startSeconds: 340 endSeconds: 430 watch: 'https://liquidgold.cc/watch/s-aixZYJG4c?t=340' heldReason: rights - ordinal: 5 text: null startSeconds: 442 endSeconds: 529 watch: 'https://liquidgold.cc/watch/s-aixZYJG4c?t=442' heldReason: rights - ordinal: 6 text: null startSeconds: 529 endSeconds: 541 watch: 'https://liquidgold.cc/watch/s-aixZYJG4c?t=529' heldReason: rights - ordinal: 7 text: null startSeconds: 583 endSeconds: 594 watch: 'https://liquidgold.cc/watch/s-aixZYJG4c?t=583' heldReason: rights - ordinal: 8 text: null startSeconds: 594 endSeconds: 642 watch: 'https://liquidgold.cc/watch/s-aixZYJG4c?t=594' heldReason: rights - ordinal: 9 text: null startSeconds: 642 endSeconds: 652 watch: 'https://liquidgold.cc/watch/s-aixZYJG4c?t=642' heldReason: rights - ordinal: 10 text: null startSeconds: 745 endSeconds: 758 watch: 'https://liquidgold.cc/watch/s-aixZYJG4c?t=745' heldReason: rights - ordinal: 11 text: null startSeconds: 800 endSeconds: 810 watch: 'https://liquidgold.cc/watch/s-aixZYJG4c?t=800' heldReason: rights - ordinal: 12 text: null startSeconds: 810 endSeconds: 853 watch: 'https://liquidgold.cc/watch/s-aixZYJG4c?t=810' heldReason: rights - ordinal: 13 text: null startSeconds: 853 endSeconds: 866 watch: 'https://liquidgold.cc/watch/s-aixZYJG4c?t=853' heldReason: rights - ordinal: 14 text: null startSeconds: 866 endSeconds: 1005 watch: 'https://liquidgold.cc/watch/s-aixZYJG4c?t=866' heldReason: rights - ordinal: 15 text: null startSeconds: 1005 endSeconds: 1016 watch: 'https://liquidgold.cc/watch/s-aixZYJG4c?t=1005' heldReason: rights - ordinal: 16 text: null startSeconds: 1016 endSeconds: 1134 watch: 'https://liquidgold.cc/watch/s-aixZYJG4c?t=1016' heldReason: rights --- # The Last Human Code Review: Building Trust in AI-Generated Code — Itamar Friedman, Qodo If you are shipping AI generated code faster than your humans can review it, Itamar Friedman's position is that you are inside the problem rather than ahead of it. He asks the room whether developers will still be reading diffs line by line by the end of the year, then reports finding two incompatible camps among engineering leaders the night before: one holding that every line must be human trusted, the other content to ship bugs and fix them quickly because velocity wins. Which camp you sit in decides what you have to build. His claim is that models stopped being the constraint. Code review benchmarks have barely moved across recent model releases, and the difference between a review that catches a real contract break and one that asks whether you considered error handling is context, not reasoning. That context is scattered across competing instruction files, differs between teams inside the same company, and largely is not written down at all, living instead in senior developers' heads and in Slack threads. Codifying it means building for two audiences at once, because the format an agent parses cleanly is not the format developers will actually maintain. The deeper version encodes the architecture itself, including which service contract broke production three months ago, so that review shifts from reading one pull request to reading a graph and noticing that three changes in flight are about to collide. Speaker info: - https://twitter.com/itamar_mar - https://www.linkedin.com/in/itamarf - https://www.qodo.ai/authors/itamar-f Timestamps: 0:00 - Where the bottleneck moved, and why code review exists 3:38 - Two camps: trust every line, or ship and fix fast 5:19 - Models are not the barrier, context is 7:03 - Context scattered across competing instruction files 8:48 - The knowledge lives in heads and in Slack 9:42 - Codifying for agents and humans at once 10:34 - Interfaces for both: rules shown, and a note to the next agent 12:19 - Fewer human comments as the readiness signal 13:11 - Encoding architecture, contracts, and past outages 14:05 - Automatic approve and block, added gradually 16:38 - Reviewing the software graph instead of the PR --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/s-aixZYJG4c-the-last-human-code-review-building-trust-in-ai-generated-code Join our upcoming AI Engineer conferences (see https://ai.engineer landing page): - AIEi Shanghai, November 5–6 — the first for meeting top Chinese labs and engineers - AIE CODE in SF, November 10–12 — the top AI Coding conference returns - AIEi Sydney, December 7–8 — the first AIEi run alongside NeurIPS 2026 in Sydney --- Canonical: https://liquidgold.cc/channel/aidotengineer/explore/video/the-last-human-code-review-building-trust-in-ai-gen-74c11726ce36 · Channel: [AI Engineer](https://liquidgold.cc/channel/aidotengineer.md) > Search chunk text is held until public-source provenance is verified. ## Document chunks > Missing: no stored document chunks are available.