CtrlK
BlogDocsLog inGet started
Tessl Logo
PluginPublic

ainativedev/aidevcon-2026-nyc

AI Native DevCon 2026 New York — practical agent skills grounded in conference talks

Quality

86%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Low

Low-risk findings worth noting

Overview
Quality
Evals
Security
Files

outline.mdskills/talk-carmont-evolving-agent-harnesses/references/

Transcript outline

Use the original cue numbers in the transcript. Promotional cues 227–237 are omitted there. This navigation aid adds no independent evidence.

TimeCuesTopic
00:00:17,174–00:02:44,7345–42Lean harness and model-release feedback loop
00:03:25,854–00:07:29,41453–109Distinguish model, harness and evaluation problems
00:07:35,294–00:09:59,934110–143Repeated verification and model behavior
00:10:21,374–00:12:09,214148–174Harness shelf life and prompt reduction report
00:12:15,774–00:14:19,104175–201Un-harnessing theories and contradictory instructions
00:14:26,174–00:16:08,874202–226Example bias and progressive disclosure
00:17:42,401–00:19:55,121249–280Remove, baseline, restore, ship and iterate
00:20:02,581–00:21:59,781281–308Evaluation ownership and outcome grading
00:22:52,841–00:24:35,881321–346Modularity, temporary scaffolding and trace feedback
00:25:17,521–00:27:09,361358–385Q&A: reward hacking and evaluation isolation
00:27:33,286–00:29:02,086391–410Q&A: tentative score range and impossible tasks
00:29:24,684–00:30:34,564414–433Experiment-oriented engineering mindset

Exact short quotations

  • “The main point here being make sure to verify your verifications.” — 00:09:48,774, cue 141.
  • “But the common theme is simplification.” — 00:15:50,194, cue 221.

Evidence limits

  • Over-80% prompt reduction is a reported team example, not a universal target.
  • Roughly 40% savings apply to some observed use cases, not every harness.
  • The 60–80% range is explicitly a tentative heuristic.
  • Trace complaints require investigation, not automatic changes to grading.
  • Current model behavior and releases are not established by the dated talk.

skills

talk-carmont-evolving-agent-harnesses

tile.json