Shipping Like Crazy

Over the last few weeks, I’ve become the product leader of a small organization, directing a team of top-notch, highly-independent engineers. We’ve shipped a whole new core service (an identity provider) and made dramatic improvements to two existing ones¹.

As you might have guessed, no, I don’t have a new job, and those engineers aren’t people. I’m commanding a fleet of Claude Code agents.

I wrote just three weeks ago about the state of agentic coding in 2025, and I feel a bit silly writing an addendum so soon. But my experience has been so positive and my productivity so high that I feel like things have changed in a meaningful way; it’s not just another moment of everyone saying the models have gotten better but then forgetting about it after a few days. I’ve been doing this for a couple of weeks and remain in awe.

This began because I subscribed to the Max tier of Claude, enabling me more usage of the best model (Opus 4.5) than anyone could ever use in their free time. Back in November and December, I saw a lot of internet discussion about how much better Claude Code had recently become, and I decided that now was the time to shell out for the higher tier.

And it’s been worth it. Will I keep paying $100 or $200 a month for this going forward? I’m not sure, but the fact that I’m even considering it is insane. But Claude Code is insanely valuable.

Product Owner and Tech Lead

In the last few weeks, I’ve shipped tens of thousands of lines of code (maybe more – who’s counting), but I’ve typed no more than a handful of them. My agentic coding workflow used to be largely hands-off, but I’d still occasionally drop into the code to figure out things that the agent was confused by, or to better understand the architecture so I could better guide the agent.

That basically hasn’t been necessary. Like the models I used before it, Opus is very good at solving clearly-defined, isolated tasks. But what sets it apart is its judgment and autonomy. I can point it toward a goal, ask it to plan the implementation, and pretty much trust that it’ll do the right thing.

This has transformed my role into some kind of combo product owner and tech lead. At the pace I’m able to ship things, I can suddenly be incredibly ambitious: I used to assume a big new feature would take me weeks of hacking in my free time, but now I can get a PR from Claude in an hour or less².

My Shifting Role

I’m still figuring out exactly what this means to me and my career. My first impression is that product and design skills are much more valuable, since I feel like my current level of technical skill is enough to guide Claude through the vast majority of features. I’ve started reading Inspired and I’d like to tackle some UX design books or trainings next.

On the other hand, I continue to feel that technical skills are valuable. Not for coding, though – only for reasoning about systems and applying technical design principles in conversations with agents.

When features take you one hour instead of ten, you can feel a bit like a superhero. What you do with these superpowers is the question that’s been on my mind a lot these last few weeks. For now, I’m just trying ship code like crazy, and produce as much useful stuff as I can think of.