Over the past few weeks during Automattic’s Radical Speed Month, Liam Sarsfield and I built an autonomous agent that goes from a project issue to a pull request. We call it Pocket Cat.

This post covers what Pocket Cat is, how it works, and what happened when we put it in the hands of our support team.

Meet Pocket Cat

All our bots share a common name for Pocket Casts: Pocket Cat. 🐈 So when it came time to name an AI agent that could autonomously pick up, investigate, and fix issues — the choice was obvious!

Here’s how it works:

  1. Someone assigns Pocket Cat to a Linear issue.
  2. The agent reads the ticket, pulls in context from internal tools, and makes a judgment call: Can I solve this?
  3. If no — it explains why, surfaces what’s missing, and asks follow-up questions.
  4. If yes — it drafts an implementation plan, writes the code, and opens a PR.

Copilot reviews the diff. Pocket Cat iterates. A human gives the final sign-off before merge.

The agent knows when to act and when to stop. That self-awareness is what makes it trustworthy enough to hand real work to.

Radical Speed Month: Two Experiments

During Automattic’s Radical Speed Month, we put Pocket Cat to the test in two ways:

  1. A brand-new Pocket Casts feature: details coming soon on the Pocket Casts blog (no spoilers!)
  2. Happiness Engineers using it to fix bugs: this is the one I want to focus on.

Support Engineers Fixing Bugs With Pocket Cat

There’s no technical barrier to entry. Using Pocket Cat is as simple as clicking a button in Linear. No IDE. No terminal. No git.

So we asked our Pocket Casts Happiness Engineers — the people who talk to customers every day and understand the pain points firsthand — to assign Pocket Cat to real bugs.

One week. Twelve issues. Three platforms.

Pocket Cat autonomously read each bug report, wrote the fix, opened the PR, and explained its reasoning.

  • iOS: 8 issues
  • Web: 3 issues
  • Desktop: 1 issue

The scorecard:

  • 7 merged
  • 3 in review
  • 2 still in draft
  • 10 of 12 PRs were self-rated high confidence by the agent

No human-written code was needed to open any of them. Just review and merge.

It’s not perfect. In one case, the agent opened a fix in the Desktop repository when the bug actually lived in a different one. And for a few issues, it decided they were too complex to tackle on its own — though it always explained why.

Why This Matters

This isn’t about replacing developers. It’s about removing the gap between knowing about a problem and fixing it.

Think about who can now kick off a fix:

  • A support specialist who just saw the same bug reported for the fifth time today
  • A designer who spotted a layout that’s off by a few pixels
  • A product manager who wants to test a quick improvement
  • A developer who’d rather review a ready-made PR than context-switch into the issue

Anyone assigns. Pocket Cat attempts. Humans review.

We still require that final human review — and that’s by design. We want to make sure:

  • The code is solid
  • The issue is actually fixed
  • Pocket Casts remains something made by humans, for humans

Pocket Cat doesn’t replace the craft. It accelerates it — and opens the door for everyone on the team to participate in shipping.