--- schema: 1 type: video title: 'Loop Engineering from First Principles — Kyle Mistele, HumanLayer' description: >- A coding agent will happily hand you a 40,000 line pull request that nobody can review and that quietly does the wrong thing. Kyle Mistele's argument is that the fix is not a better prompt but a better loop, borrowed from control theory: a thermostat senses the error between where a system is and where you want it, emits a control signal, and measures again, over and over. Infrastructure as code already approximates this. The point is to design agent loops the same way, so each iteration makes a small, readable change you can actually verify, instead of one giant diff you have to trust. The working example is migrating a codebase one procedure at a time. A sensor, often just Grep or a structural search, finds the smallest unmigrated piece, a controller picks what to work on next, and an actuator agent makes the change against golden patterns defined by hand, gated by deterministic CI like a single loop iteration in CircleCI. The loop tracks its own PRs in version control, refuses to stack a new change while an earlier one is still open, and keeps improving the code incrementally, even while the team is away. Speaker info: - https://x.com/0xBlacklight - https://www.linkedin.com/in/kyle-mistele - https://blacklight.sh Timestamps: 0:00 - Introduction: the 40,000 line PR problem 1:55 - Why more code is not the goal 3:37 - Is the generated code any good? 4:43 - Control loops from control theory 5:49 - Infrastructure as code and Ralph loops 7:06 - Applying control loops to coding 8:35 - Migrating a codebase one procedure at a time 10:40 - Tracking the loop in version control 12:35 - The actuator agent and golden patterns 13:51 - Wiring the loop into CI 15:44 - Avoiding stacked PRs and scaling the controller --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/xIt_mTQp6mY-loop-engineering-from-first-principles 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/watch/xIt_mTQp6mY' markdown: 'https://liquidgold.cc/watch/xIt_mTQp6mY.md' llms: 'https://liquidgold.cc/watch/xIt_mTQp6mY/llms.txt' parent: 'https://liquidgold.cc/channel/aidotengineer.md' channel: aidotengineer updated: '2026-10-24T06:30:15.205Z' language: null children: [] videoId: xIt_mTQp6mY durationSeconds: 1077 publishedAt: '2026-07-25T20:41:40.000Z' searchChunks: - ordinal: 0 text: null startSeconds: 39 endSeconds: 117 watch: 'https://liquidgold.cc/watch/xIt_mTQp6mY?t=39' heldReason: rights - ordinal: 1 text: null startSeconds: 39 endSeconds: 117 watch: 'https://liquidgold.cc/watch/xIt_mTQp6mY?t=39' heldReason: rights - ordinal: 2 text: null startSeconds: 128 endSeconds: 228 watch: 'https://liquidgold.cc/watch/xIt_mTQp6mY?t=128' heldReason: rights - ordinal: 3 text: null startSeconds: 239 endSeconds: 294 watch: 'https://liquidgold.cc/watch/xIt_mTQp6mY?t=239' heldReason: rights - ordinal: 4 text: null startSeconds: 294 endSeconds: 305 watch: 'https://liquidgold.cc/watch/xIt_mTQp6mY?t=294' heldReason: rights - ordinal: 5 text: null startSeconds: 305 endSeconds: 351 watch: 'https://liquidgold.cc/watch/xIt_mTQp6mY?t=305' heldReason: rights - ordinal: 6 text: null startSeconds: 351 endSeconds: 361 watch: 'https://liquidgold.cc/watch/xIt_mTQp6mY?t=351' heldReason: rights - ordinal: 7 text: null startSeconds: 361 endSeconds: 427 watch: 'https://liquidgold.cc/watch/xIt_mTQp6mY?t=361' heldReason: rights - ordinal: 8 text: null startSeconds: 438 endSeconds: 515 watch: 'https://liquidgold.cc/watch/xIt_mTQp6mY?t=438' heldReason: rights - ordinal: 9 text: null startSeconds: 515 endSeconds: 527 watch: 'https://liquidgold.cc/watch/xIt_mTQp6mY?t=515' heldReason: rights - ordinal: 10 text: null startSeconds: 527 endSeconds: 645 watch: 'https://liquidgold.cc/watch/xIt_mTQp6mY?t=527' heldReason: rights - ordinal: 11 text: null startSeconds: 645 endSeconds: 657 watch: 'https://liquidgold.cc/watch/xIt_mTQp6mY?t=645' heldReason: rights - ordinal: 12 text: null startSeconds: 657 endSeconds: 764 watch: 'https://liquidgold.cc/watch/xIt_mTQp6mY?t=657' heldReason: rights - ordinal: 13 text: null startSeconds: 764 endSeconds: 774 watch: 'https://liquidgold.cc/watch/xIt_mTQp6mY?t=764' heldReason: rights - ordinal: 14 text: null startSeconds: 774 endSeconds: 838 watch: 'https://liquidgold.cc/watch/xIt_mTQp6mY?t=774' heldReason: rights - ordinal: 15 text: null startSeconds: 849 endSeconds: 951 watch: 'https://liquidgold.cc/watch/xIt_mTQp6mY?t=849' heldReason: rights - ordinal: 16 text: null startSeconds: 849 endSeconds: 951 watch: 'https://liquidgold.cc/watch/xIt_mTQp6mY?t=849' heldReason: rights - ordinal: 17 text: null startSeconds: 951 endSeconds: 961 watch: 'https://liquidgold.cc/watch/xIt_mTQp6mY?t=951' heldReason: rights - ordinal: 18 text: null startSeconds: 961 endSeconds: 1077 watch: 'https://liquidgold.cc/watch/xIt_mTQp6mY?t=961' heldReason: rights --- # Loop Engineering from First Principles — Kyle Mistele, HumanLayer A coding agent will happily hand you a 40,000 line pull request that nobody can review and that quietly does the wrong thing. Kyle Mistele's argument is that the fix is not a better prompt but a better loop, borrowed from control theory: a thermostat senses the error between where a system is and where you want it, emits a control signal, and measures again, over and over. Infrastructure as code already approximates this. The point is to design agent loops the same way, so each iteration makes a small, readable change you can actually verify, instead of one giant diff you have to trust. The working example is migrating a codebase one procedure at a time. A sensor, often just Grep or a structural search, finds the smallest unmigrated piece, a controller picks what to work on next, and an actuator agent makes the change against golden patterns defined by hand, gated by deterministic CI like a single loop iteration in CircleCI. The loop tracks its own PRs in version control, refuses to stack a new change while an earlier one is still open, and keeps improving the code incrementally, even while the team is away. Speaker info: - https://x.com/0xBlacklight - https://www.linkedin.com/in/kyle-mistele - https://blacklight.sh Timestamps: 0:00 - Introduction: the 40,000 line PR problem 1:55 - Why more code is not the goal 3:37 - Is the generated code any good? 4:43 - Control loops from control theory 5:49 - Infrastructure as code and Ralph loops 7:06 - Applying control loops to coding 8:35 - Migrating a codebase one procedure at a time 10:40 - Tracking the loop in version control 12:35 - The actuator agent and golden patterns 13:51 - Wiring the loop into CI 15:44 - Avoiding stacked PRs and scaling the controller --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/xIt_mTQp6mY-loop-engineering-from-first-principles 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/watch/xIt_mTQp6mY · 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.