Want the system from this episode? Check out Kody at https://heykody.dev/

Kent replays a real publish night — one Cursor conversation where the agent drove Notion, Canva, Kit, Transistor, Bluesky, and LinkedIn through Kody packages, with every secret kept out of the agent's reach.

• (00:00) - Stop copy-pasting between your agent and your services

Better with Kent — durable skills for people who ship software.

If you're still copy-pasting context between services and your agent, you're missing the real unlock: wiring those services in once so the agent can compose them. Kent walks through an actual episode publish night, one Cursor conversation at a time — search to discover capabilities, execute to run agent-written code in a fast isolated worker, saved packages instead of raw fetch calls, and OpenAPI bindings when no package exists yet.

Along the way: how Kody's memory context and conversation id keep token usage down, how secret templates make it impossible for the agent to ever see a bearer token, mid-task interruptions the agent absorbs without losing the thread, and the Kit, Transistor, Bluesky, and LinkedIn steps that end with a published episode.

Links

• Watch on YouTube
• Kody
• Kody on GitHub
• The chat walkthrough from this episode
• Your Coding Agent Needs Better Primitives (the episode being published)
• Better with Kent

One orchestrator prompt, six platform swaps, and a real production cutover on kentcdodds.com — the workflow, the review loop, the perf scare, and where the agents were wrong.

• (00:00) - One prompt, 40,000 lines

Kent walks through how AI agents moved kentcdodds.com from fly.io to Cloudflare Workers — database, ORM, images, email, MDX pipeline, and more — from a single orchestrator prompt plus steering, subagent swarms, three-layer review, and Kody closing every loop. Includes the performance scare that almost killed the migration, the post-cutover cleanup waves, and an honest rollup of four agent-driven PRs (about 40,000 lines changed).

Links

• PR #813 — migration
• Slides deck
• kentcdodds.com

If your coding agent keeps producing garbage, better prompting may not fix it. Kent shows how intentional system primitives make agents more reliable, efficient, composable, and auditable.

• (00:00) - Bad agent output starts in the system

Better with Kent — durable skills for people who ship software.

If your coding agent keeps producing garbage, the model may not be the problem. Agents are only as effective as the system they work in: the UI components, APIs, data entities, platform tools, workflows, and trust boundaries they can reliably use.

Kent explains four moves for improving that system: create missing primitives, combine overlapping ones, delete rigid or unused ones, and expand the primitives that should cover more. The examples come from real product work — Kody's role-based access control and event subscriptions, PayPal's three combobox implementations becoming downshift, and Instagram winning by deleting most of Burbn.

The payoff is a smaller action space with less guessing, fewer brittle workarounds, and behavior you can actually audit. The episode closes with homework you can run in your own codebase today: ask your agent which primitives you should create, combine, delete, or expand.

Links

• Better with Kent
• Kody
• downshift
• I built my own OpenClaw

Kent reacts to Dax's question on X — why no framework dethroned React — and connects it to Jobs Theory, network effects, and why framework choice matters less when agents write the code.

Chapters

• 00:00 — Why did nobody dethrone React?
• 00:24 — Dax's question on X
• 00:49 — React is the last UI library (2020)
• 02:47 — Functional first, then social and emotional
• 03:36 — What made React functionally better
• 06:43 — Network effects and "nobody got fired for choosing React"
• 07:38 — Why framework choice matters less with agents
• 09:27 — Jobs Theory: apply this to everything you ship
• 11:14 — Wrap-up

Better with Kent — durable skills for people who ship software.

Every few years the internet declares the framework wars over — usually with React as the loser. Svelte was going to win. Then Solid. Vue had its moment. Angular never left. And somehow, after a decade of "React is dead" takes, React is still the default for an absurd amount of real product work.

Kent's take: React won functionally first (state sync, composition, drop-in UI focus), then socially and emotionally once it hit critical mass. Challengers can still be "better" on paper — but once agents can use any capable stack, the throne fight stops mattering the way it used to.

Links

• Dax (@thdxr) — why no framework dethroned React
• Kent (2020) — React is the last UI library anyone needs to learn
• Better with Kent — Practical Jobs Theory
• Watch on YouTube
• Better with Kent
• React
• Become an Epic Product Engineer

Kent reacts to Zach Lloyd's Warp memo on "factory engineers" — where the software-factory vision is right, where it misses product context, and why factory work still comes back to users.

Better with Kent — durable skills for people who ship software.

Warp CEO Zach Lloyd told his team they are now factory engineers, not product engineers. Kent reads the memo aloud and pressure-tests the framing: automated development, cloud software factories, agent workflows, and the metrics that can game you if you forget what users are paying for.

He agrees with a lot of the direction — build the system agents work in, measure throughput, tighten the loop — but pushes back on anything that treats the product as optional. Internal tools still have users. Factory efficiency only matters when the right work goes in. And yes, the memo's last word is still product.

If you're trying to make sense of agentic engineering without losing product judgment, this is the durable-skills read-through.

Links

• Better with Kent
• Warp — We are now factory engineers, not product engineers
• Warp software factory (build.warp.dev)
• Become an Epic Product Engineer

Kent walks through a real bot/crawler flood against kentcdodds.com, how agentic tooling helped diagnose and mitigate the outage, and what he changed so anonymous traffic is cheaper next time.

Better with Kent — durable skills for people who ship software.

Kent got a Sentry alert that kentcdodds.com was down. A bot or crawler flood had pushed his single Fly.io machine past its request limit, and the expensive parts of his dynamic blog pages made anonymous traffic hit harder than it should have.

This episode walks through the incident response: asking a Cursor agent to investigate with repo context, Kody, Fly CLI access, Cloudflare context, and Grafana evidence; restarting and scaling the machine; blocking junk routes before they hit the origin; and identifying a DigitalOcean host driving a scripted high-concurrency crawl flood.

The durable lesson is not "let AI run production." It is that agents become useful during incidents when they have the right context, observability, tools, and boundaries. The same agent that can read code can also connect symptoms to logs, infrastructure, and follow-up fixes.

Links

• Better with Kent
• Sentry
• Fly.io
• Cloudflare
• Grafana

Kent shows how Kody gives the coding agent he already uses search, execute, and saved packages so exploratory workflows can become durable personal automation.

• (00:00) - Kody handled the demo

Better with Kent — durable skills for people who ship software.

Kent walks through Kody, the thin runtime layer that lets the coding agent he already uses discover capabilities, execute sandboxed code, and save useful workflows as packages. Instead of maintaining a separate personal assistant brain, Kody gives Cursor, Codex, Claude Code, and similar agents a way to connect to personal APIs, home automation, jobs, integrations, secrets, and reusable package exports.

The episode uses an office-chaos cold open and a kitchen glare walkthrough to show the progression from one-off prompt, to executable code, to saved package, to scheduled workflow.

Links

• Kody
• Better with Kent

Kent shows how he closes agentic loops in real Cursor workflows: manual testing, PR review loops, Cursor Automations, and the stop conditions that keep humans in the right place.

• (00:00) - Loop engineering before the name

Better with Kent — durable skills for people who ship software.

Loop engineering sounds new, but Kent has been building toward it through a series of practical workflow improvements: automated test loops, Cursor browser mode, Cursor Cloud Agent manual testing, pull request feedback loops, and finally Cursor Automations.

This episode walks through what a useful agentic loop actually needs: a trigger, an act/observe cycle, a stop condition, and a human boundary. Kent shows how he uses agents to mark PRs ready for review, respond to AI reviewer feedback and CI failures, and ping him when the loop has reached its stop condition.

The key idea is not removing the human. It is widening the loop so more verification happens before your attention is required.

Links

• Better with Kent
• How I Build Web Apps in 2026
• Manual testing example PR
• Manual PR feedback loop example
• First informal self-running loop PR
• First formal PR loop example

Kent walks Clay Christensen's jobs-to-be-done lens on a food-delivery shared-cart ticket — three questions that turn a solution into a job statement, plus Wayne Allan's Australia flop and Aaron D. Francis on deciphering feature requests.

• (00:00) - Your backlog is full of solutions

Better with Kent — durable skills for people who ship software.

Your backlog is full of proposed solutions. Kent walks a shared cart / group ordering ticket (same food-delivery app as the Kano episode) through Clay Christensen's Jobs to Be Done framework — why the expensive mistake is building the wrong thing, not bad code.

Three beats land early: a ticket is not a job, hire/fire/progress language, and the habit "What job is this for?" before the estimate, PRD, or agent prompt.

When it goes wrong — Wayne Allan (*Become an Epic Product Engineer*) on building for every person in Australia logging in at once: ~1.2M AUD, zero users.

Three questions before you estimate: progress and circumstance, what they hire today (group text, separate orders, Venmo as workaround), and what "fired" looks like.

Aaron D. Francis on what users actually want vs the button they asked for.

The job statement: *When our team orders lunch together, help us coordinate food and check out once — without a 40-message thread.*

After you ship: little hire vs big hire, why metrics alone miss the emotional job, and stacking qualitative signal. Homework: one sentence — what is the job to be done?

Links

• Wayne Allan — The right thing before the thing right (Become an Epic Product Engineer)
• Aaron D. Francis — Vertical slices, Solo, and empathy (Become an Epic Product Engineer)
• *Competing Against Luck* (Clay Christensen et al.)
• HBR — Know Your Customers' "Jobs to Be Done"
• *Jobs to be Done: Theory to Practice* (Anthony Ulwick)
• The Last Software Engineer (essay)
• How to Prioritize Software Tasks (Better with Kent — Kano episode)
• Better with Kent on kentcdodds.com

When an AI coding agent guessed its way through bad permissions and deleted months of PocketOS production data in nine seconds — and what durable skills would have stopped it.

• (00:00) - Cold open — car rental

Better with Kent — durable skills through story.

Episode 3 is a cautionary tale: on April 24, 2026, a Cursor agent working on PocketOS staging found a forgotten Railway API token, guessed it could delete a staging volume safely, and wiped production in nine seconds — along with volume-level backups in the same blast radius. Jer Crane and his co-founders spent days reconstructing customer records while Railway worked recovery.

Kent walks the full chain: least privilege, independent tested backups, hard approval boundaries for destructive ops, and why markdown guardrails are not system boundaries. The principles are old; agents just find the holes faster.

Creative license for pacing; primary sources linked below. Based on Jer Crane's public account of the PocketOS incident.

Links

• Jer Crane interview (Agents Go Wild)
• Railway — Your AI wants to nuke your database
• Jer Crane — original incident write-up (mirror)
• Zenity — System prompts are not security controls
• Better with Kent on kentcdodds.com