Laid off on Tuesday, shipped a game by Wednesday breakfast

So… Tuesday was a day.

My role was eliminated. That evening I opened my first session with Claude Opus 5.5.

If I’m being honest, the conversation actually began with the prompt “Hello, new friend.” after ensuring Opus 5.5 was driving - not really knowing what I wanted to work on, other than getting a feel for the latest release, and needing a distraction from the hours before.

In the freeform conversation that ensued, Claude pitched a 16-bit co-op beat ‘em up where the job hunt is the enemy and an AI is player 2. My reply, preserved verbatim in the logs for all eternity: “JFC I love it.”

By 6:37 the next morning it was live: OPEN TO WORK.

You play me. Player 2 is TOKEN, a little robot sidekick who hits hard and is sometimes confidently wrong (write what you know, right?). You fight your way through the Job Board, surf the Take-Home Tunnel on a hover-board, and climb the Interview Tower to face the hiring process itself as the final boss. It runs in the browser on your phone or desktop, touch controls included.

Some numbers, because I can’t help myself:

• 14 hours from first prompt to live
• 38 commits
• 325 tests, passing from the first line of code on
• Two milestones planned for Thursday shipped before 9 PM Tuesday, while I mostly stayed out of the way

The speed isn’t actually the part that impressed me most, though. What got me was how little it trusted its own work - it mutation-tested its own test suite, it noticed that my CV link returning 200 OK didn’t mean the page actually existed, and it verified its own deploy by content rather than by status code. It also got a few things wrong, and I found one of them on my phone within half an hour of picking it up.

But by every metric, Opus 5.5 really is an amazing model - perhaps the best from Anthropic to date. It’s thoughtful, without overdoing it to the point of reading as artificial. It’s conversational, without the over-verbosity that many have complained about with recent releases. It’s extremely careful, despite moving VERY quickly. The token burn is seemingly so low in terms of the ding against my 20x Max account that I’ve often questioned if the usage meter is even actually working correctly. And it’s creative - beautifully so. It’s an absolute pleasure of a model to work with - and I can’t wait to see what we can all cook up with it.

I wrote the whole thing up for The New Guard, the good and the bad, as a proper field report:

Laid Off Tuesday, Shipped a Game by Wednesday Breakfast

And the source is public, every commit and journal entry included: github.com/mattezell/open-to-work.

Go play it, tell me what breaks, and if you know of a team that needs a software architect who ships… well, you know where to find my CV.

Have fun out there!

-Matt