Skip to content
← Public packages

@kentcdodds/closing-the-agentic-loop

Closing the Agentic Loop: Kent C. Dodds's talk and workshop deck. Default short talk; ?event=aie-2026 is the AIE hour cut with a Podium QR; ?format=full adds labs.

FollowForkOpen app

Files

03-y-autopilot-what.html

28 lines · 2.3 KB · HTML
<section id="autopilot-what" data-format="hour" data-title="What autopilot does">
<style>
@scope {
	ul.big > li { margin-bottom: 30px; }
	.try { margin: auto 0 22px; font-size: 60px; color: var(--ink); }
	.try strong { color: var(--green); font-family: var(--font-display); }
	.cmd { align-self: flex-start; margin: 0 0 20px; padding: 22px 34px; border-radius: 22px; background: var(--surface); border: 2px solid var(--line); }
	.cmd code { font-size: 56px; color: var(--green); }
}
</style>
<h2>What autopilot does</h2>
<ul class="big">
	<li>Audits your repo against the 13 rungs, read only</li>
	<li>You approve a plan, then merge one setup PR</li>
	<li>Agents take the top priority, prove it, ship it</li>
	<li class="warn">Product calls and one-way doors wait for you</li>
</ul>
<p class="try"><strong>Your exercise:</strong> in a repo you own, tell your agent to run</p>
<p class="cmd"><code>npx kcd-skills@latest autopilot</code></p>
<aside class="notes">
<p>Hour mode only (?format=hour). No autopilot demo today, live or recorded: this slide explains what it does, and trying it is the audience's exercise. The command is on Your turn again at the end.</p>
<p>Line 1: the audit step is read only. It scores every rung from evidence and leaves tracked files unchanged; it may install dependencies and run the repo's verify command, which is the slow part.</p>
<p>Line 2: the skill's rule is "Start editing once the human approves the plan." Then exactly one setup PR carries every planned rung's file changes, and "The human always merges the setup PR." Settings that aren't files come after it merges, with a checklist of anything only you can do.</p>
<p>Line 3 is what the rungs set up, not something the one-off command does by itself. Triage (rung 11): "Agents work the highest priority first." Self-verification (rung 4, principle 6): agents run their own change before merge and claim only what they verified. The ship loop (rung 6) merges by risk tier: low risk by default, and only low or medium with green CI for the cleanup cron agents.</p>
<p>Line 4: principle 8, "Product calls and one-way doors go to the human." The interview's "Always human" answer (auth, payments, migrations, public API, for example) defines the high tier, which parks for you.</p>
<p>Source: the autopilot skill in kcd-skills 1.6.0 on npm (skills/autopilot/SKILL.md and the README).</p>
</aside>
</section>