BEGIN:VCALENDAR
VERSION:2.0
PRODID:Data::ICal 0.24
BEGIN:VEVENT
DESCRIPTION:   'Title: Nix Knows When the Agent Is Wrong\n   Tags: Nix Vega
 s Community | Creator Talk/Panel\n   When: Saturday\, Aug 8\, 14:00 - 14:3
 0 PDT\n   Where: LVCCW Level 1 Hall 4 1310 (Nix Vegas Community) - [1]Map\
 n\n   Description:\n\n   I built a tool to help maintainers patch nixpkgs 
 CVEs faster. But I\n   optimized the wrong thing.\n\n   The tool - Vulnpat
 ch\, is a dashboard. It pulls together CVE\n   intelligence\, triages by w
 hat is actually being exploited\, gives the\n   maintainer the context the
 y need and gets out of the way. The\n   assumption underneath it was that 
 the maintainer is the bottleneck:\n   speed up the workflow and packages g
 et patched faster.\n\n   I no longer believe that assumption. I was alread
 y building Vulnpatch\n   to automate the remediation work when Mythos was 
 announced\, and it\n   undercut the premise. Building a product to do what
  a frontier model\n   could now do on its own was the wrong problem to be 
 solving.\n\n   The difficult part was never automating the work. It is mak
 ing an\n   agent's output something a volunteer on the security team can a
 ctually\n   trust. The work is the evidence: which package is actually aff
 ected\,\n   which upstream commit fixes it\, what the new hash is\, whethe
 r it still\n   builds\, whether the tests still pass and whether a maintai
 ner has any\n   reason to trust it. And that is before the non-trivial cas
 es: the\n   patches that are not simple version bumps or hash updates.\n\n
    This is a talk about that shift. I will discuss Vulnpatch and Trace\,\n
    an agent-owned nixpkgs fork that produces signed\, reproducible CVE\n  
  evidence bundles instead of drive-by pull requests. And I want to make\n 
   one argument I think is worth taking seriously: Nix is unusually\n   wel
 l-suited to AI agents because it provides strong automated\n   feedback. R
 eproducible builds and passthru.tests give agents an\n   objective verifie
 r for many changes\, whereas most repositories offer\n   much weaker valid
 ation.\n\n   Patches are cheap. Proof is not. I will show what it takes to
  make an\n   agent's work auditable enough to be worth a maintainer's time
  and I\n   hope\, to persuade maintainers that agents belong in their work
 flow.\n   The workflows we built for human-only contribution will not surv
 ive\n   contact with agents unchanged.\n\n   SpeakerBio:  Jason Odoom\n\n 
   "Thing Doer"\n\n   '\n\n   1. #LVCCW_Level1_Hall4\n\n\n
DTEND:20260808T213000Z
DTSTART:20260808T210000Z
LOCATION:Nix Vegas Community - LVCCW Level 1 Hall 4 1310 (Nix Vegas Communi
 ty)
SUMMARY:Nix Knows When the Agent Is Wrong
END:VEVENT
END:VCALENDAR
