Superpowers have changed how I write software (mostly)

I’ve been having lots of fun the past few weeks chatting with people at work about my current agentic coding setup. Since late 2025, most of my coding work has been done with the assistance of Claude Code’s agentic skills feature. In particular, the Superpowers skills created by Jesse Vincent have been gamechangers!

In this article, I’ll share my workflow for thinking about and creating software, using Superpowers–and why I feel they’re worth checking out if you’ve had poor or mixed results with AI-assisted or agentic coding tools in the past.

(Agentic skills are increasingly table stakes and now a feature in many tools, including OpenAI’s Codex and open source alternatives like Crush and OpenCode. I’ve not tried these agents myself, but the general workflow should apply in them, too.)

The Superpowers workflow

The Superpowers skill is actually several skills, which can be used individually. But for the most part, I use its highest-level tools to think through problems and ideas, craft plans based on that thinking, and then do the work to make that plan happen.

The coding session kicks off with a brainstorm. Here, I’ll begin with providing up-front context to the agent. Then, I’ll have a conversation with the agent, to clarify and refine its understanding of the request–and more important, my own understanding!

For simpler tasks, the agent may suggest just doing the thing. For larger or more complex projects, the agent transitions to the write plan skill. The plan is a human-readable Markdown file, so it’s reviewable, or I can implement it myself, or I can hand it off to the agent.

Assuming I am tasking the agent with the work, it’ll use the execute plan to make it happen! I’ve got a couple of options:

• Work side-by-side with the agent, pausing for review and feedback in incremental steps. I really like this mode when I’m using the tools to help me understand a new-to-me code base or concept! In fact, I’ve explicitly asked for plans that call out teaching moments along the way. It’s a nice, interactive alternative to more traditional learning methods.
• Permit the agent to outsource the work to subagents, which are tasked with writing tests, implementing code, reviewing the work, documenting, and, committing for later review. Each agent is prompted for the specific type of work it’s doing.

I always recommend the first approach to people who are new to agentic coding, or for work that they couldn’t do totally on their own if they had to–then trust the agents to do the work end-to-end as comfort levels grow.

Using Superpowers beyond code

I’ve recently begun a systematic organization and cleanup of my Obsidian notes–something that’s always been a struggle for me. Through brainstorming, I put together a plan to work through my notes, then (hopefully) keep them tidy. I kicked off the session acknowledging that my current system didn’t work, and that I’m open to change. It was really neat watching the agent notice patterns in my notes and suggest baby steps to get me toward better notes. I’m still early in this process and may write more about it at some point.

It can chew through tokens

Be advised–as I tackled problems of increased scope with Superpowers, I found myself hitting the token allotment on my Claude Code Pro account pretty quickly. I’ve since upgraded to a Max account and haven’t run into limits. It’s worth the cost for me, but I can respect that it may be a non-starter for many. (I am interested in testing superpowers with an agent running against a locally-running model, but that is a someday project.)

Critical thinking is your real superpower

Here’s the thing: the more thought and consideration you put into a problem up front, the more likely your agent will generate net positive results. It’s the same as working a vague feature spec or ticket: Sure, you could run with it, but if you don’t have the full context, there’s a decent chance you’ll come back with a result that doesn’t match the request.

Good software developers know this inherently, but it takes practice (and a good dose of humility!) to get there. Superpowers’s brainstorming skill gives us a safe space to build these habits and strengths–to push back on fuzzy ideas, ask But why? countless times like a toddler, work through constraints, and consider alternatives.

It’s still up to us, as humans, to decide whether the resulting co-created plan solves the problem, and just as important, that the problem actually needs solving. It’s still on us to look at the resulting code and sign off that it does the thing it’s supposed to do and meets expectations around maintainability, security, and performance.

That’s critical thinking. No agent is doing it for you. But brainstorming with Superpowers can help you build and strengthen your own critical thinking muscles–then flex them the next time a vague user request comes in.