--- schema: 1 type: video title: >- Building Blocks for Uber’s Software Factory— Uday Kiran Medisetty & Adam Huda, Uber description: >- More than 70% of pull requests at Uber now come from local or cloud agents, and lines of code per engineer has doubled year over year. Uday Kiran Medisetty walks through the six pieces of infrastructure underneath that, and the constraint shaping all of them shows up in the first one: every model call in the company goes through a single gateway doing Spire identity, redaction of 20 plus PII types, and five specialized safety models, with that entire guardrail budget held under 100 milliseconds. It carries 100 million requests a day across 800 projects, each one attributable to a caller, team, and project. Most of the rest is about not drowning agents in tokens. An MCP gateway crawls internal APIs into MCP servers, then projects them into a CLI so responses stay out of the context window, which cut fleetwide token use by more than 40%. A skills marketplace holds 2,500 entries behind lint checks and automated review, running 20,000 executions a day. A context graph of 150 node and edge types and 40 million entries replaces the 20 to 30 separate systems agents used to crawl for basic ownership and dependency facts. Adam Huda then takes one feature end to end, from a Slack thread to a draft PR that deliberately stops short of CI so that validation happens in the inner loop, comparing simulator screenshots against Figma specs before anything reaches a build queue. His closing point is that the bottleneck has moved to whether a thing should be built at all. Speaker info: Uday Kiran Medisetty: - https://x.com/udaykiran - https://www.linkedin.com/in/udaykiran/ Adam Huda: - https://x.com/hudaman - https://www.linkedin.com/in/thinktopdown/ - https://adamhuda.com Timestamps: 0:00 - The numbers: 70% of PRs, twice the code per engineer 1:26 - The model gateway, and a 100 millisecond guardrail budget 3:51 - MCP gateway, and cutting the token tax 5:48 - Dev pods, agentified into pre-provisioned balloon pods 7:18 - A managed skills marketplace 8:39 - The context graph, 40 million entries 10:19 - Cortana across Slack, CLI, and web 11:34 - One feature end to end, starting from a Slack thread 13:28 - Minion, and stopping short of CI on purpose 16:01 - Maintenance as a managed loop 17:29 - The bottleneck is now whether you should build it --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/17-YSUHo6Lk-agentic-sdlc-at-uber-building-blocks-for-uber-s-software-factory 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/17-YSUHo6Lk' markdown: 'https://liquidgold.cc/watch/17-YSUHo6Lk.md' llms: 'https://liquidgold.cc/watch/17-YSUHo6Lk/llms.txt' parent: 'https://liquidgold.cc/channel/aidotengineer.md' channel: aidotengineer updated: '2026-10-24T05:01:47.522Z' language: null children: [] videoId: 17-YSUHo6Lk durationSeconds: 1106 publishedAt: '2026-08-21T13:00:06.000Z' searchChunks: - ordinal: 0 text: null startSeconds: 12 endSeconds: 22 watch: 'https://liquidgold.cc/watch/17-YSUHo6Lk?t=12' heldReason: rights - ordinal: 1 text: null startSeconds: 22 endSeconds: 91 watch: 'https://liquidgold.cc/watch/17-YSUHo6Lk?t=22' heldReason: rights - ordinal: 2 text: null startSeconds: 22 endSeconds: 91 watch: 'https://liquidgold.cc/watch/17-YSUHo6Lk?t=22' heldReason: rights - ordinal: 3 text: null startSeconds: 102 endSeconds: 238 watch: 'https://liquidgold.cc/watch/17-YSUHo6Lk?t=102' heldReason: rights - ordinal: 4 text: null startSeconds: 102 endSeconds: 238 watch: 'https://liquidgold.cc/watch/17-YSUHo6Lk?t=102' heldReason: rights - ordinal: 5 text: null startSeconds: 251 endSeconds: 351 watch: 'https://liquidgold.cc/watch/17-YSUHo6Lk?t=251' heldReason: rights - ordinal: 6 text: null startSeconds: 251 endSeconds: 351 watch: 'https://liquidgold.cc/watch/17-YSUHo6Lk?t=251' heldReason: rights - ordinal: 7 text: null startSeconds: 363 endSeconds: 440 watch: 'https://liquidgold.cc/watch/17-YSUHo6Lk?t=363' heldReason: rights - ordinal: 8 text: null startSeconds: 451 endSeconds: 522 watch: 'https://liquidgold.cc/watch/17-YSUHo6Lk?t=451' heldReason: rights - ordinal: 9 text: null startSeconds: 532 endSeconds: 623 watch: 'https://liquidgold.cc/watch/17-YSUHo6Lk?t=532' heldReason: rights - ordinal: 10 text: null startSeconds: 634 endSeconds: 700 watch: 'https://liquidgold.cc/watch/17-YSUHo6Lk?t=634' heldReason: rights - ordinal: 11 text: null startSeconds: 712 endSeconds: 809 watch: 'https://liquidgold.cc/watch/17-YSUHo6Lk?t=712' heldReason: rights - ordinal: 12 text: null startSeconds: 820 endSeconds: 963 watch: 'https://liquidgold.cc/watch/17-YSUHo6Lk?t=820' heldReason: rights - ordinal: 13 text: null startSeconds: 820 endSeconds: 963 watch: 'https://liquidgold.cc/watch/17-YSUHo6Lk?t=820' heldReason: rights - ordinal: 14 text: null startSeconds: 820 endSeconds: 963 watch: 'https://liquidgold.cc/watch/17-YSUHo6Lk?t=820' heldReason: rights - ordinal: 15 text: null startSeconds: 963 endSeconds: 974 watch: 'https://liquidgold.cc/watch/17-YSUHo6Lk?t=963' heldReason: rights - ordinal: 16 text: null startSeconds: 1056 endSeconds: 1106 watch: 'https://liquidgold.cc/watch/17-YSUHo6Lk?t=1056' heldReason: rights --- # Building Blocks for Uber’s Software Factory— Uday Kiran Medisetty & Adam Huda, Uber More than 70% of pull requests at Uber now come from local or cloud agents, and lines of code per engineer has doubled year over year. Uday Kiran Medisetty walks through the six pieces of infrastructure underneath that, and the constraint shaping all of them shows up in the first one: every model call in the company goes through a single gateway doing Spire identity, redaction of 20 plus PII types, and five specialized safety models, with that entire guardrail budget held under 100 milliseconds. It carries 100 million requests a day across 800 projects, each one attributable to a caller, team, and project. Most of the rest is about not drowning agents in tokens. An MCP gateway crawls internal APIs into MCP servers, then projects them into a CLI so responses stay out of the context window, which cut fleetwide token use by more than 40%. A skills marketplace holds 2,500 entries behind lint checks and automated review, running 20,000 executions a day. A context graph of 150 node and edge types and 40 million entries replaces the 20 to 30 separate systems agents used to crawl for basic ownership and dependency facts. Adam Huda then takes one feature end to end, from a Slack thread to a draft PR that deliberately stops short of CI so that validation happens in the inner loop, comparing simulator screenshots against Figma specs before anything reaches a build queue. His closing point is that the bottleneck has moved to whether a thing should be built at all. Speaker info: Uday Kiran Medisetty: - https://x.com/udaykiran - https://www.linkedin.com/in/udaykiran/ Adam Huda: - https://x.com/hudaman - https://www.linkedin.com/in/thinktopdown/ - https://adamhuda.com Timestamps: 0:00 - The numbers: 70% of PRs, twice the code per engineer 1:26 - The model gateway, and a 100 millisecond guardrail budget 3:51 - MCP gateway, and cutting the token tax 5:48 - Dev pods, agentified into pre-provisioned balloon pods 7:18 - A managed skills marketplace 8:39 - The context graph, 40 million entries 10:19 - Cortana across Slack, CLI, and web 11:34 - One feature end to end, starting from a Slack thread 13:28 - Minion, and stopping short of CI on purpose 16:01 - Maintenance as a managed loop 17:29 - The bottleneck is now whether you should build it --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/17-YSUHo6Lk-agentic-sdlc-at-uber-building-blocks-for-uber-s-software-factory 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/17-YSUHo6Lk · 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.