--- schema: 1 type: video title: 'Generative UI... in Python? — Jeremiah Lowin, Prefect' description: >- Because only the agent can reach an MCP server, a naive file upload tool makes the agent retype the file into the server character by character. Jeremiah Lowin calls it the world's most expensive copy paste operation, and it is the problem MCP apps exist to remove. Lowin is founder and CEO of Prefect and the author of FastMCP. MCP apps, a protocol extension added this year, let a tool result skip the agent and arrive at the user as HTML, CSS, and JavaScript: a real interface a person can click. That solves one problem and creates another, because FastMCP's users are overwhelmingly Python engineers inside enterprises, and shipping React from Python is not something he would pretend to do. The way out was constraining the problem. His users are not building consumer facing branded products; they share and collect information, which means tables, forms, and charts. Prefab is the result, a Python DSL where nesting context managers builds the interface structure and every component is a class that renders as a polished one. Reactive variables give client side interactivity with no JavaScript. The pipeline runs Python to a declarative representation to a JSON protocol to a React app, and Lowin is blunt that the JSON is the actual point: a serializable UI is one an agent can generate, receive, or edit. The Python DSL, he says, was an accident he noticed later. Two details land. The Prefab documentation is rendered entirely in Prefab, editable in place. And a UI streamed as Python turned out roughly seventy percent smaller than the same UI as JSON, so they now send Python over the wire and convert it in a sandbox. Speaker info: - https://x.com/jlowin - https://www.linkedin.com/in/jlowin/ - https://jlowin.dev Timestamps: 0:00 - MCP apps, and bypassing the agent 3:20 - Shipping UIs to an audience of Python engineers 4:45 - Constraining the problem to tables, forms, charts 5:43 - Composing a front end rather than building one 6:37 - The DSL, and the JSON in the middle 8:54 - Docs rendered in the thing they document 10:13 - Interactive tools, and a chart beside a table 12:56 - Full apps and the expensive copy paste problem 15:15 - Generative UI, and Python beating JSON on the wire --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/Krzs8GeiWTc-youtube-1adaa0e5b93ef60154ca 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/Krzs8GeiWTc' markdown: 'https://liquidgold.cc/watch/Krzs8GeiWTc.md' llms: 'https://liquidgold.cc/watch/Krzs8GeiWTc/llms.txt' parent: 'https://liquidgold.cc/channel/aidotengineer.md' channel: aidotengineer updated: '2026-10-24T03:48:32.822Z' language: null children: [] videoId: Krzs8GeiWTc durationSeconds: 1058 publishedAt: '2026-09-10T16:30:07.000Z' searchChunks: - ordinal: 0 text: null startSeconds: 23 endSeconds: 206 watch: 'https://liquidgold.cc/watch/Krzs8GeiWTc?t=23' heldReason: rights - ordinal: 1 text: null startSeconds: 23 endSeconds: 206 watch: 'https://liquidgold.cc/watch/Krzs8GeiWTc?t=23' heldReason: rights - ordinal: 2 text: null startSeconds: 23 endSeconds: 206 watch: 'https://liquidgold.cc/watch/Krzs8GeiWTc?t=23' heldReason: rights - ordinal: 3 text: null startSeconds: 219 endSeconds: 287 watch: 'https://liquidgold.cc/watch/Krzs8GeiWTc?t=219' heldReason: rights - ordinal: 4 text: null startSeconds: 287 endSeconds: 297 watch: 'https://liquidgold.cc/watch/Krzs8GeiWTc?t=287' heldReason: rights - ordinal: 5 text: null startSeconds: 355 endSeconds: 399 watch: 'https://liquidgold.cc/watch/Krzs8GeiWTc?t=355' heldReason: rights - ordinal: 6 text: null startSeconds: 399 endSeconds: 410 watch: 'https://liquidgold.cc/watch/Krzs8GeiWTc?t=399' heldReason: rights - ordinal: 7 text: null startSeconds: 410 endSeconds: 538 watch: 'https://liquidgold.cc/watch/Krzs8GeiWTc?t=410' heldReason: rights - ordinal: 8 text: null startSeconds: 410 endSeconds: 538 watch: 'https://liquidgold.cc/watch/Krzs8GeiWTc?t=410' heldReason: rights - ordinal: 9 text: null startSeconds: 410 endSeconds: 538 watch: 'https://liquidgold.cc/watch/Krzs8GeiWTc?t=410' heldReason: rights - ordinal: 10 text: null startSeconds: 538 endSeconds: 549 watch: 'https://liquidgold.cc/watch/Krzs8GeiWTc?t=538' heldReason: rights - ordinal: 11 text: null startSeconds: 627 endSeconds: 778 watch: 'https://liquidgold.cc/watch/Krzs8GeiWTc?t=627' heldReason: rights - ordinal: 12 text: null startSeconds: 627 endSeconds: 778 watch: 'https://liquidgold.cc/watch/Krzs8GeiWTc?t=627' heldReason: rights - ordinal: 13 text: null startSeconds: 789 endSeconds: 925 watch: 'https://liquidgold.cc/watch/Krzs8GeiWTc?t=789' heldReason: rights - ordinal: 14 text: null startSeconds: 789 endSeconds: 925 watch: 'https://liquidgold.cc/watch/Krzs8GeiWTc?t=789' heldReason: rights - ordinal: 15 text: null startSeconds: 925 endSeconds: 936 watch: 'https://liquidgold.cc/watch/Krzs8GeiWTc?t=925' heldReason: rights - ordinal: 16 text: null startSeconds: 936 endSeconds: 1058 watch: 'https://liquidgold.cc/watch/Krzs8GeiWTc?t=936' heldReason: rights --- # Generative UI... in Python? — Jeremiah Lowin, Prefect Because only the agent can reach an MCP server, a naive file upload tool makes the agent retype the file into the server character by character. Jeremiah Lowin calls it the world's most expensive copy paste operation, and it is the problem MCP apps exist to remove. Lowin is founder and CEO of Prefect and the author of FastMCP. MCP apps, a protocol extension added this year, let a tool result skip the agent and arrive at the user as HTML, CSS, and JavaScript: a real interface a person can click. That solves one problem and creates another, because FastMCP's users are overwhelmingly Python engineers inside enterprises, and shipping React from Python is not something he would pretend to do. The way out was constraining the problem. His users are not building consumer facing branded products; they share and collect information, which means tables, forms, and charts. Prefab is the result, a Python DSL where nesting context managers builds the interface structure and every component is a class that renders as a polished one. Reactive variables give client side interactivity with no JavaScript. The pipeline runs Python to a declarative representation to a JSON protocol to a React app, and Lowin is blunt that the JSON is the actual point: a serializable UI is one an agent can generate, receive, or edit. The Python DSL, he says, was an accident he noticed later. Two details land. The Prefab documentation is rendered entirely in Prefab, editable in place. And a UI streamed as Python turned out roughly seventy percent smaller than the same UI as JSON, so they now send Python over the wire and convert it in a sandbox. Speaker info: - https://x.com/jlowin - https://www.linkedin.com/in/jlowin/ - https://jlowin.dev Timestamps: 0:00 - MCP apps, and bypassing the agent 3:20 - Shipping UIs to an audience of Python engineers 4:45 - Constraining the problem to tables, forms, charts 5:43 - Composing a front end rather than building one 6:37 - The DSL, and the JSON in the middle 8:54 - Docs rendered in the thing they document 10:13 - Interactive tools, and a chart beside a table 12:56 - Full apps and the expensive copy paste problem 15:15 - Generative UI, and Python beating JSON on the wire --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/Krzs8GeiWTc-youtube-1adaa0e5b93ef60154ca 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/Krzs8GeiWTc · 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.