How I Learned to Stop Worrying and Love Claude

When these AI coding tools first came out, I was pretty skeptical of them, especially considering how earlier models would make silly mistakes and write lower quality code than I could write myself. I found myself utilizing them a lot more though in 2025 with the release of Opus 4 and Sonnet 4, and I started trying to figure out a good process for working with them. In the latter half of 2025, I started to experiment with Cursor, using its Plan mode with Sonnet and then using Composer (codenamed cheetah) to execute on those plans, and I noticed that that specific workflow had promising results. With the release of Gemini 3, I found even better results within Cursor and realized that it was a decent enough model for building front-ends, and this kind of set off a lightbulb for me and I began to think about what else I could build with these tools.

Goodbye decision paralysis

My biggest weakness when it came to side projects over the last few years is that I would get decision paralysis when it came to building. I would fret over details, whether it’s the architecture, design, marketing, or whatever and just not actually build the thing. After working a full day of work, I’d also find my brain too fried to actually want to write more code, usually wanting time to decompress and relax. I had a realization through using these AI tools that I could take these documents of ideas and architectures and actually use these AI tools to help me execute on them. The AI tools became a means to an end for me, a tool I could use to actually bring these ideas to life and iterate on, rather than just obsessing over the details and trying to perfect them in my head.

How I learned to stop worrying and love Claude

Once I had this realization, all of my excuses for not shipping were eliminated. I started with using Gemini 3 for my side projects, starting with Oru, a job board as a service application that I had been putting off for a while. A few weeks after Gemini 3, Opus 4.5 was released and was a HUGE game changer in terms of model quality. It wasn’t until Opus that I switched back to using Claude Code as my main coding tool, and the quality of the output was just so much better that I was addicted to the feeling of coding again. It felt like I was back in college, just shipping random things for the hell of it.

Since Opus 4.5 was released, I’ve found myself working on projects I had neglected or forgotten about, even the silly ones. The latest iteration of this very website was built using Claude Code, complete with a custom CMS for uploading photographs and drafting blog posts on the go, all deployed to Cloudflare. Nowadays, if I have an idea for a project, I immediately go to my laptop and start exploring how it can be built with Claude Code or Cursor. The barrier to entry is basically eliminated.

Building tools to build with agents

Given the amount of time I was spending in Claude Code, I naturally started building tools to improve my workflows. Grove arose from my need for tracking the work of parallelized agents and allowing them to run apps under randomly selected ports for testing in the browser without stepping on each other. Tasuku was born out of testing beads and not liking the complexity and weight, so I decided to not just build a task management system for agents, but also a means for them to continually reinforce knowledge and learnings as they are building to help improve output over time. These AI tools keep evolving, and the need for tooling around them will evolve too. If you’re experiencing a pain point while working with these tools, build something to fix it and dogfood it into your workflow.

The future of building

Honestly, I thought AI would affect my love for programming but if anything, it’s just reignited the passion even more and I’m excited for the future. As a programmer, programming is a means to an end for solving problems, and I view these various AI tools as another tool. Every now and then, AI still does stupid things and you have to steer it in the right direction, but that is becoming less frequent as the models improve.

Recently, Anthropic released Opus 4.6 and OpenAI released GPT 5.3 Codex, both of which are extremely impressive models and perform pretty well. And we’re only in February, meaning that there is still time for Google to release Gemini 4 and plenty of updates for other frontier models. I don’t know what the state of the ecosystem will look like by the end of 2026, but I cannot wait to see and continue to use this new technology. If you’re still skeptical, I would encourage you to give it a shot for yourself and see how you feel afterwards. Figure out what works best for you and just start building. I promise you won’t regret it.