--- schema: 1 type: video title: 'Stop Prompting — Greg Pstrucha, Sentry' description: >- You correct your agent. It says "You're absolutely right." Five minutes later, after compaction, it makes the same mistake again. Greg Pstrucha, who works on the infrastructure behind Seer, Sentry's debugging agent, argues you should stop re-prompting your coding agent and start codifying the rules you keep repeating. Agents have no memory, so the same elementary mistakes come back: overly defensive code, useless tests, endpoints nobody needs. Tests, strict types and especially linters turn those corrections into deterministic checks, and with agents, writing a custom lint rule is now cheap. He shows real examples from Sentry: lint rules that keep endpoints in sync with the OpenAPI schema, and a long game of whack-a-mole to keep agent skills accurate. For qualitative judgment, he uses written policies and a "Grumpy Engineer" review loop, and warns that agents will game any number you hand them, like test coverage. In this talk: • Why re-prompting fails, and why lint rules are now cheap to write • Mining your transcripts and PR reviews for rules worth enforcing • Examples from Sentry: API schema checks and linting agent skills • Policies and review agents for what linters can't catch, and why metrics get gamed SPEAKER Greg Pstrucha, Staff Software Engineer, Sentry LinkedIn: https://www.linkedin.com/in/greg-pstrucha/ LINKS Sentry Seer: https://sentry.io/product/seer/ Sentry: https://sentry.io Sentry on X: https://x.com/sentry CHAPTERS 0:00 Intro 0:38 Who still reads AI code? 1:13 The code is bad 1:58 "You're absolutely right" 2:23 Stop prompting, start codifying 2:58 Tests, strict types, linters 4:08 Why lint rules are cheap now 5:28 Mine your transcripts for rules 5:58 Custom lint rules 6:43 Example: keeping APIs in sync 7:52 Example: linting agent skills 9:17 Whack-a-mole with the agent 10:37 Tools: ESLint, ast-grep, Taskless 11:27 When linters aren't enough 12:22 Policies and the Grumpy Engineer 14:07 Beware of gamed metrics 15:22 Call to action Recorded at the AI Engineer World's Fair 2026 in San Francisco. Subscribe for more talks from the engineers building with AI. AI Engineer: https://ai.engineer YouTube: https://www.youtube.com/@aiDotEngineer X: https://x.com/aiDotEngineer LinkedIn: https://www.linkedin.com/company/aidotengineer/ #CodingAgents #CodeQuality #AIEngineer canonical: 'https://liquidgold.cc/watch/E3KbFLAGD6A' markdown: 'https://liquidgold.cc/watch/E3KbFLAGD6A.md' llms: 'https://liquidgold.cc/watch/E3KbFLAGD6A/llms.txt' parent: 'https://liquidgold.cc/channel/aidotengineer.md' channel: aidotengineer updated: '2026-10-23T22:19:28.944Z' language: null children: [] videoId: E3KbFLAGD6A durationSeconds: 956 publishedAt: '2026-10-09T14:00:20.000Z' searchChunks: - ordinal: 0 text: null startSeconds: 87 endSeconds: 119 watch: 'https://liquidgold.cc/watch/E3KbFLAGD6A?t=87' heldReason: rights - ordinal: 1 text: null startSeconds: 119 endSeconds: 129 watch: 'https://liquidgold.cc/watch/E3KbFLAGD6A?t=119' heldReason: rights - ordinal: 2 text: null startSeconds: 129 endSeconds: 151 watch: 'https://liquidgold.cc/watch/E3KbFLAGD6A?t=129' heldReason: rights - ordinal: 3 text: null startSeconds: 163 endSeconds: 184 watch: 'https://liquidgold.cc/watch/E3KbFLAGD6A?t=163' heldReason: rights - ordinal: 4 text: null startSeconds: 184 endSeconds: 196 watch: 'https://liquidgold.cc/watch/E3KbFLAGD6A?t=184' heldReason: rights - ordinal: 5 text: null startSeconds: 196 endSeconds: 249 watch: 'https://liquidgold.cc/watch/E3KbFLAGD6A?t=196' heldReason: rights - ordinal: 6 text: null startSeconds: 260 endSeconds: 329 watch: 'https://liquidgold.cc/watch/E3KbFLAGD6A?t=260' heldReason: rights - ordinal: 7 text: null startSeconds: 329 endSeconds: 341 watch: 'https://liquidgold.cc/watch/E3KbFLAGD6A?t=329' heldReason: rights - ordinal: 8 text: null startSeconds: 374 endSeconds: 406 watch: 'https://liquidgold.cc/watch/E3KbFLAGD6A?t=374' heldReason: rights - ordinal: 9 text: null startSeconds: 417 endSeconds: 476 watch: 'https://liquidgold.cc/watch/E3KbFLAGD6A?t=417' heldReason: rights - ordinal: 10 text: null startSeconds: 488 endSeconds: 565 watch: 'https://liquidgold.cc/watch/E3KbFLAGD6A?t=488' heldReason: rights - ordinal: 11 text: null startSeconds: 577 endSeconds: 643 watch: 'https://liquidgold.cc/watch/E3KbFLAGD6A?t=577' heldReason: rights - ordinal: 12 text: null startSeconds: 643 endSeconds: 654 watch: 'https://liquidgold.cc/watch/E3KbFLAGD6A?t=643' heldReason: rights - ordinal: 13 text: null startSeconds: 654 endSeconds: 688 watch: 'https://liquidgold.cc/watch/E3KbFLAGD6A?t=654' heldReason: rights - ordinal: 14 text: null startSeconds: 698 endSeconds: 742 watch: 'https://liquidgold.cc/watch/E3KbFLAGD6A?t=698' heldReason: rights - ordinal: 15 text: null startSeconds: 742 endSeconds: 754 watch: 'https://liquidgold.cc/watch/E3KbFLAGD6A?t=742' heldReason: rights - ordinal: 16 text: null startSeconds: 754 endSeconds: 847 watch: 'https://liquidgold.cc/watch/E3KbFLAGD6A?t=754' heldReason: rights - ordinal: 17 text: null startSeconds: 858 endSeconds: 926 watch: 'https://liquidgold.cc/watch/E3KbFLAGD6A?t=858' heldReason: rights - ordinal: 18 text: null startSeconds: 926 endSeconds: 956 watch: 'https://liquidgold.cc/watch/E3KbFLAGD6A?t=926' heldReason: rights --- # Stop Prompting — Greg Pstrucha, Sentry You correct your agent. It says "You're absolutely right." Five minutes later, after compaction, it makes the same mistake again. Greg Pstrucha, who works on the infrastructure behind Seer, Sentry's debugging agent, argues you should stop re-prompting your coding agent and start codifying the rules you keep repeating. Agents have no memory, so the same elementary mistakes come back: overly defensive code, useless tests, endpoints nobody needs. Tests, strict types and especially linters turn those corrections into deterministic checks, and with agents, writing a custom lint rule is now cheap. He shows real examples from Sentry: lint rules that keep endpoints in sync with the OpenAPI schema, and a long game of whack-a-mole to keep agent skills accurate. For qualitative judgment, he uses written policies and a "Grumpy Engineer" review loop, and warns that agents will game any number you hand them, like test coverage. In this talk: • Why re-prompting fails, and why lint rules are now cheap to write • Mining your transcripts and PR reviews for rules worth enforcing • Examples from Sentry: API schema checks and linting agent skills • Policies and review agents for what linters can't catch, and why metrics get gamed SPEAKER Greg Pstrucha, Staff Software Engineer, Sentry LinkedIn: https://www.linkedin.com/in/greg-pstrucha/ LINKS Sentry Seer: https://sentry.io/product/seer/ Sentry: https://sentry.io Sentry on X: https://x.com/sentry CHAPTERS 0:00 Intro 0:38 Who still reads AI code? 1:13 The code is bad 1:58 "You're absolutely right" 2:23 Stop prompting, start codifying 2:58 Tests, strict types, linters 4:08 Why lint rules are cheap now 5:28 Mine your transcripts for rules 5:58 Custom lint rules 6:43 Example: keeping APIs in sync 7:52 Example: linting agent skills 9:17 Whack-a-mole with the agent 10:37 Tools: ESLint, ast-grep, Taskless 11:27 When linters aren't enough 12:22 Policies and the Grumpy Engineer 14:07 Beware of gamed metrics 15:22 Call to action Recorded at the AI Engineer World's Fair 2026 in San Francisco. Subscribe for more talks from the engineers building with AI. AI Engineer: https://ai.engineer YouTube: https://www.youtube.com/@aiDotEngineer X: https://x.com/aiDotEngineer LinkedIn: https://www.linkedin.com/company/aidotengineer/ #CodingAgents #CodeQuality #AIEngineer Canonical: https://liquidgold.cc/watch/E3KbFLAGD6A · Channel: [AI Engineer](https://liquidgold.cc/channel/aidotengineer.md) > Search chunk text is held until public-source provenance is verified. ## Document chunks --- #### Chunk 0: Document excerpt (0:00 to 0:59) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 1: Document excerpt (0:59 to 1:59) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 2: Document excerpt (1:59 to 2:31) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 3: Document excerpt (2:31 to 3:04) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 4: Document excerpt (3:04 to 4:09) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 5: Document excerpt (4:09 to 5:29) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 6: Document excerpt (5:29 to 6:03) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 7: Document excerpt (6:03 to 6:46) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 8: Document excerpt (6:46 to 7:56) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 9: Document excerpt (7:56 to 9:25) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 10: Document excerpt (9:25 to 10:43) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 11: Document excerpt (10:43 to 11:28) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 12: Document excerpt (11:28 to 12:22) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 13: Document excerpt (12:22 to 14:07) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 14: Document excerpt (14:07 to 15:26) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 15: Document excerpt (15:26 to 15:56) > Held: transcript rights do not permit this chunk's text.