<?xml version="1.0" encoding="utf-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
    <channel>
        <title>fighting slop with slop — Vaibhav Gupta, Boundary</title>
        <link>https://video.ut0pia.org/videos/watch/b5e64260-a88a-49df-b90b-378520376e16</link>
        <description>You cannot tell great engineers what to do, and you increasingly cannot tell what an agent did either, so Vaibhav Gupta's answer is to fight slop with slop. At Boundary the team turns the same cheap, sloppy generation loose as a tool: agents that run constantly over the transcripts of other agents, flagging hallucinations, spotting which tool calls produced errors, and comparing which approaches produced fewer. He pairs that with hard invariants, the design docs, rules, and CLI checks that do not change for months and tell you exactly where a codebase stops converging, so the messy detection layer sits on top of something stable. The deeper move is to attack the foundational layer from first principles. Instead of trusting generated code, he leans on type systems that make whole classes of mistakes impossible: types get inferred without you writing them, a division by zero is guaranteed to be handled or the code will not build, and there are no silent unknowns left for an agent to guess at. That is the bet behind BAML, which lets you work across Python, TypeScript, or Rust with strong boundaries around each function so an agent can move fast inside walls it cannot breach. His closing challenge is to go build these sloppy tools yourself and constrain the systems underneath them, because that is what actually wins the war on slop. Speaker info: https://x.com/vaibcode, https://www.linkedin.com/in/vaigup, https://www.youtube.com/@boundaryml, Timestamps: 0:00 - Fighting slop with slop 0:27 - Code reviews and invariants 2:08 - Building rules that don't change 3:00 - Design docs with notifications 4:03 - CLI tools that catch where things break 5:08 - Agents reading agent transcripts 5:57 - Detecting hallucinations and errors 9:24 - Attacking the foundational layer 11:03 - Execution traces from first principles 14:26 - Type safe tools across platforms 15:42 - Making whole error classes impossible 17:57 - BAML across languages 20:37 - Go build the sloppy tools</description>
        <lastBuildDate>Fri, 31 Jul 2026 17:03:23 GMT</lastBuildDate>
        <docs>https://validator.w3.org/feed/docs/rss2.html</docs>
        <generator>PeerTube - https://video.ut0pia.org</generator>
        <image>
            <title>fighting slop with slop — Vaibhav Gupta, Boundary</title>
            <url>https://video.ut0pia.org/lazy-static/avatars/0287a09a-aae7-4840-9843-b416426e7046.webp</url>
            <link>https://video.ut0pia.org/videos/watch/b5e64260-a88a-49df-b90b-378520376e16</link>
        </image>
        <copyright>All rights reserved, unless otherwise specified in the terms specified at https://video.ut0pia.org/about and potential licenses granted by each content's rightholder.</copyright>
        <atom:link href="https://video.ut0pia.org/feeds/video-comments.xml?videoId=b5e64260-a88a-49df-b90b-378520376e16" rel="self" type="application/rss+xml"/>
    </channel>
</rss>