--- schema: 1 type: video title: 'From Signal to PR: Anatomy of a Self-Improving Agent — Jason Lopatecki, Arize' description: >- Instead of getting paged at midnight and starting to dig, you wake up to an issue that has already been investigated: the traces pulled, the root cause found, and a pull request with the fix waiting for review. That is what Arize built with Signal, and Jason Lopatecki walks through the anatomy of it. The unlock is boring and specific: traces on a filesystem. A skill pulls the relevant production traces and logs down as files into the repo, right next to the code, sometimes ten megabytes of them, because coding harnesses like Claude Code are magical with files and hopeless with a dashboard. From there the agent has the exact code path the software took, not a guess among a million branches, and can produce a real fix. You pick the harness, the sandbox, and the skills, and Arize can run it inside your VPC, because companies like Uber and Booking will not point production systems at an external model. The deeper shift is that observability stops being a dashboard you click and becomes the smoke a system throws off for agents to read, which is why you now log and trace ten times more, not less. He is honest about the limits: a one line fix is the easy case, bigger fixes still need a human to drive, and the job moves from responder to reviewer. Speaker info: - https://www.linkedin.com/in/jason-lopatecki-9509941/ - https://arize.com/author/jason-lopatecki/ Timestamps: 0:00 - Arize, its agent Alyx, and why v1 sucked 1:36 - Observability is changing: from dashboards to telemetry for agents 2:55 - The goal: systems that fix themselves 4:14 - Inverting the loop: the agent investigates first 6:08 - Traces on a filesystem, the key unlock 7:23 - From your laptop to sandboxes 8:13 - A real fix: the Alyx stream canceled bug 9:39 - Why you should trace ten times more 11:10 - Product demo: Signal, AX, and Phoenix 13:09 - Sandboxes, VPC, and why customers won't call out 16:20 - Q&A: why not just point Claude Code at your data? 18:04 - Q&A: where do the evals come in? --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/9HbzAWnKbo4-from-signal-to-pr-anatomy-of-a-self-improving-agent 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/from-signal-to-pr-anatomy-of-a-self-improving-agent-5b860bf2f12b markdown: >- https://liquidgold.cc/channel/aidotengineer/video/from-signal-to-pr-anatomy-of-a-self-improving-agent-5b860bf2f12b.md llms: null parent: 'https://liquidgold.cc/channel/aidotengineer.md' channel: aidotengineer updated: '2026-10-24T06:36:04.922Z' language: null children: [] videoId: 9HbzAWnKbo4 durationSeconds: 1236 publishedAt: '2026-07-24T20:15:32.000Z' searchChunks: - ordinal: 0 text: null startSeconds: 108 endSeconds: 175 watch: 'https://liquidgold.cc/watch/9HbzAWnKbo4?t=108' heldReason: rights - ordinal: 1 text: null startSeconds: 186 endSeconds: 263 watch: 'https://liquidgold.cc/watch/9HbzAWnKbo4?t=186' heldReason: rights - ordinal: 2 text: null startSeconds: 186 endSeconds: 263 watch: 'https://liquidgold.cc/watch/9HbzAWnKbo4?t=186' heldReason: rights - ordinal: 3 text: null startSeconds: 273 endSeconds: 371 watch: 'https://liquidgold.cc/watch/9HbzAWnKbo4?t=273' heldReason: rights - ordinal: 4 text: null startSeconds: 273 endSeconds: 371 watch: 'https://liquidgold.cc/watch/9HbzAWnKbo4?t=273' heldReason: rights - ordinal: 5 text: null startSeconds: 371 endSeconds: 381 watch: 'https://liquidgold.cc/watch/9HbzAWnKbo4?t=371' heldReason: rights - ordinal: 6 text: null startSeconds: 381 endSeconds: 447 watch: 'https://liquidgold.cc/watch/9HbzAWnKbo4?t=381' heldReason: rights - ordinal: 7 text: null startSeconds: 458 endSeconds: 493 watch: 'https://liquidgold.cc/watch/9HbzAWnKbo4?t=458' heldReason: rights - ordinal: 8 text: null startSeconds: 583 endSeconds: 596 watch: 'https://liquidgold.cc/watch/9HbzAWnKbo4?t=583' heldReason: rights - ordinal: 9 text: null startSeconds: 596 endSeconds: 673 watch: 'https://liquidgold.cc/watch/9HbzAWnKbo4?t=596' heldReason: rights - ordinal: 10 text: null startSeconds: 800 endSeconds: 986 watch: 'https://liquidgold.cc/watch/9HbzAWnKbo4?t=800' heldReason: rights - ordinal: 11 text: null startSeconds: 800 endSeconds: 986 watch: 'https://liquidgold.cc/watch/9HbzAWnKbo4?t=800' heldReason: rights - ordinal: 12 text: null startSeconds: 800 endSeconds: 986 watch: 'https://liquidgold.cc/watch/9HbzAWnKbo4?t=800' heldReason: rights - ordinal: 13 text: null startSeconds: 996 endSeconds: 1084 watch: 'https://liquidgold.cc/watch/9HbzAWnKbo4?t=996' heldReason: rights - ordinal: 14 text: null startSeconds: 1095 endSeconds: 1236 watch: 'https://liquidgold.cc/watch/9HbzAWnKbo4?t=1095' heldReason: rights - ordinal: 15 text: null startSeconds: 1095 endSeconds: 1236 watch: 'https://liquidgold.cc/watch/9HbzAWnKbo4?t=1095' heldReason: rights --- # From Signal to PR: Anatomy of a Self-Improving Agent — Jason Lopatecki, Arize Instead of getting paged at midnight and starting to dig, you wake up to an issue that has already been investigated: the traces pulled, the root cause found, and a pull request with the fix waiting for review. That is what Arize built with Signal, and Jason Lopatecki walks through the anatomy of it. The unlock is boring and specific: traces on a filesystem. A skill pulls the relevant production traces and logs down as files into the repo, right next to the code, sometimes ten megabytes of them, because coding harnesses like Claude Code are magical with files and hopeless with a dashboard. From there the agent has the exact code path the software took, not a guess among a million branches, and can produce a real fix. You pick the harness, the sandbox, and the skills, and Arize can run it inside your VPC, because companies like Uber and Booking will not point production systems at an external model. The deeper shift is that observability stops being a dashboard you click and becomes the smoke a system throws off for agents to read, which is why you now log and trace ten times more, not less. He is honest about the limits: a one line fix is the easy case, bigger fixes still need a human to drive, and the job moves from responder to reviewer. Speaker info: - https://www.linkedin.com/in/jason-lopatecki-9509941/ - https://arize.com/author/jason-lopatecki/ Timestamps: 0:00 - Arize, its agent Alyx, and why v1 sucked 1:36 - Observability is changing: from dashboards to telemetry for agents 2:55 - The goal: systems that fix themselves 4:14 - Inverting the loop: the agent investigates first 6:08 - Traces on a filesystem, the key unlock 7:23 - From your laptop to sandboxes 8:13 - A real fix: the Alyx stream canceled bug 9:39 - Why you should trace ten times more 11:10 - Product demo: Signal, AX, and Phoenix 13:09 - Sandboxes, VPC, and why customers won't call out 16:20 - Q\&A: why not just point Claude Code at your data? 18:04 - Q\&A: where do the evals come in? --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/9HbzAWnKbo4-from-signal-to-pr-anatomy-of-a-self-improving-agent 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/from-signal-to-pr-anatomy-of-a-self-improving-agent-5b860bf2f12b · 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.