12 Months of Claude - Puzzmo Blog

I try to commit to things. My relationship with the craft of programming is significantly more intense than most of the people I have worked with in my career. This isn’t a slight to others, a more diverse set of interests makes for more well-rounded people and there’s a lot of things to do as a human in a lifetime! My commitment to the craft comes with a cost - I am extremely wary of adding dependents and taking on responsibilities which do not give me maximal time and space to further the work on my craft.

The reason for being dependent-phobic is that the effort takes time. Unbelievable amounts of time. Since I started to commit

to programming as a craft about 13 years ago, I have programmed almost every day for somewhere between 8 to 10 hours. I have devoted tens of thousands of hours to understanding and contributing back to each ecosystem I’ve relied on: Ruby, iOS Native, Node, Browsers and Server Infra over that time. Those hours are based on one simple foundational concept which I grasped right at the beginning of my career: Every day I build on the work and knowledge of past me. So, any extra work I put in today gives me the chance to build upon this further tomorrow.

My last 12 months of using Claude Code has really shaken those foundations, because I think at heart, it allows for others to have access to the skills you can gain from that commitment, without putting in the time.

I find it both very exciting, and deeply epochal.

As this is the third in a series on using Claude Code: first, 1 week, second, 6 weeks. You can opt to skip them, but I will briefly get you up to speed, feel free to jump the next few paragraphs till you hit a horizontal rule to continue.

I’m Orta, one of the co-founders of Puzzmo (where we make daily web games with interesting systems around them, think Wordle meets Fortnite). Prior to that I worked on the TypeScript compiler team at Microsoft doing the odd compiler bug/feature but mostly working on docs and web infra. I have a serious backlog of open source contributions.

I lead a team of engineers here at Puzzmo, who over time had a varied amount of willingness to use or experiment with LLMs for their daily programming. I have used GitHub Copilot since it was a wee baby only in Microsoft and have a world of respect for the team working on it - I debated working on that instead of founding Puzzmo!

I found Copilot underwhelming on the TypeScript compiler, but very effective at guessing the end of my sentence when working in the fledgling Puzzmo codebase. Then this year, I explored Cursor and found myself very impressed at Cursor’s ability to infer the rest of the paragraph.

Then Claude Code came out, and completely changed what it meant to be a programmer. I found myself being able to simultaneously ship features and architectural refactors at the same time by using multiple clones. Maintenance/refactors which typically took substantial amounts of time and resources became commonplace everyday PRs as I flew through ~1,100 pull requests to Puzzmo since I started using Claude.

The list of changes from the first 6 weeks is formidable for an actually fully staffed team of engineers.

Interestingly, I found it very hard to quantify the change in a concrete metric like Pull Requests, commits or lines of code changed. I will re-explore this.

One final note before I get started, these posts sometimes end up outside of the programming ecosystem - so if you are deeply pessimistic about LLMs and their consequences, so am I! I don’t want to just put my head in the sand and pretend we are still in a pre-Claude Code world though. This stuff should be understood and discussed among domain experts.

LLMs as the “killer app” of This Generation#

I’ve seen many technical fads come and go: ‘ChatBots’, ‘Metaverse’, ‘Edutainment’, ‘Crypto’, ‘AR/VR’, ‘Uber for X’, ‘Apps’ etc. I try to understand their underlying ‘why is this happening now?’ and ‘what tech underpins this?’ but I absolutely bet against all of them.

To me, there’s not been many ‘Killer Apps’, ones which literally change how you interact with the world. The most recent prior ‘Killer App’ which really impressed me is the mix of GPS, Google Maps and the smartphone. That was some transformative tech. It’s not to say there aren’t other great bits of technology, but the idea that we could effectively never get lost planet-wide must have been unfathomable to previous generations.

I think LLMs, Reasoning Loops and Code is the next ‘Killer App’ - it’s not reached accessibility for everyone yet but almost everyone I’ve met who has genuinely engaged with it comes out changed. When I consider:

The complexity of actually understanding how a computer works, how far away...