← Public packages
@kentcdodds/the-last-software-engineer
The Last Software Engineer: Kent C. Dodds's React Summit US 2026 talk deck. Judgment and product engineering in an AI era.
Files
<section id="system" class="chat-slide" data-title="Understand / design the system" data-steps="2">
<h2 data-m="rise">Understand / design the system</h2>
<div class="chat" data-m="rise" data-d="200">
<div class="lights" aria-hidden="true"><i></i><i></i><i></i></div>
<div class="bar"><img class="avatar" src="c/img/avatar-cole.webp" alt=""><span class="name">Cole</span><span class="role">Coder</span></div>
<div class="screen" data-until="1">
<div class="msg me" style="--i:1">Is it absolutely necessary that we keep the workers <mark>keyed by the user ID</mark>? If two users have an agent that has the exact same code for execute, can we not just reuse that worker?</div>
<div class="msg them" style="--i:2">Makes total sense ... You're not missing a secret reason we must key workers by user for isolation.</div>
</div>
<div class="screen" data-at="1" data-until="2">
<div class="msg them" style="--i:0">Also: a worker's identity includes the deploy version, so <mark>every platform deploy starts every worker over</mark>. Narrowing the key to only what changes behavior fixes that.</div>
<div class="msg me" style="--i:1">To be clear my goal here is primarily to <mark>cut my own costs</mark> and then hopefully we can pass those cost savings on to users through higher entitlements or reduced pricing.</div>
</div>
<div class="screen" data-at="2">
<div class="msg me" style="--i:0">...this is a pretty impactful change. I'm fine with the agent shipping it and deploying it autonomously but I want it to do <mark>rigorous manual testing</mark> of the preview deploy.</div>
<div class="msg them" style="--i:1">Shipped: deploys no longer reset workers. Testing showed sharing one worker across users isn't safe yet, so the per-user key stays. User pricing is unchanged for now.</div>
</div>
<div class="input" aria-hidden="true"><span class="circle"><svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round"><path d="M12 5v14M5 12h14"/></svg></span><span class="ph" data-chrome>Message Cole</span><span class="circle"><svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><rect x="9" y="3" width="6" height="11" rx="3"/><path d="M5.5 11a6.5 6.5 0 0 0 13 0M12 17.5V21"/></svg></span><span class="circle send"><svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2.4" stroke-linecap="round" stroke-linejoin="round"><path d="M12 19V5M5.5 11.5 12 5l6.5 6.5"/></svg></span></div>
</div>
<aside class="notes">
<p>Third habit: understand and design the system. I didn't write this code, but I understood the system well enough to ask a pointed question: why are workers keyed by user ID? If two users run identical code, why not reuse the worker?</p>
<p>[click] That question surfaced something better: every platform deploy was starting every worker over, because the deploy version was part of the key. And I was explicit about the goal: cut my costs first, then pass savings on to users.</p>
<p>[click] And I set the risk level. Ship it autonomously, but do rigorous manual testing of the preview deploy. The testing found that sharing workers across users wasn't safe yet, so that part didn't ship. The deploy fix did. My original idea was half wrong, and the system design caught it. You don't need to write the code, but you need to understand the system well enough to ask the right question and set the right guardrails.</p>
</aside>
</section>