The Exhausting Reality of AI Agent Supervision
If you’re working with AI coding assistants, you know the pattern all too well. Every session starts the same way:
“Use TDD.” “Write tests first.” “Check for dependencies.” “Follow the established patterns.”
It’s not that the AI doesn’t understand these instructions—it does. But the next time you start a new feature or switch context, you’re saying it all over again. The 50th time you type “use test-driven development,” you start to wonder: Am I the developer, or am I a human linter for AI behavior?
The cognitive overhead is real. You’re not just solving problems; you’re constantly course-correcting, reminding, and supervising. It’s like having a brilliant but forgetful intern who needs the same onboarding every single day.
What if you could encode these instructions once and have your AI agents follow them automatically—every time, without prompting?
Enter Superpowers: Systematic Workflows, Not Supervision
Recently, I discovered Jesse Vincent’s (obra) post about Superpowers on his blog, and it fundamentally changed how I think about working with AI coding assistants.
The core insight is deceptively simple: Skills are executable knowledge. They’re not just documentation or guidelines—they’re instructions that Claude Code agents must follow when they apply to the task at hand.
Superpowers is a plugin system for Claude Code that encodes proven development workflows into “skills.” When a skill exists for a task, Claude automatically uses it. No more repeating yourself. No more babysitting. The agent knows the process and follows it systematically.
Here’s what makes this revolutionary: Jesse didn’t just create a tool—he created a system where agents aren’t just capable of following best practices, they’re compelled to follow them. The discipline is encoded, not hoped for.
Think about that for a moment. How many times have you wished your AI assistant would just remember to use TDD? Or automatically set up git worktrees for parallel development? Or ask clarifying questions before diving into implementation?
With Superpowers, those behaviors aren’t suggestions—they’re the default.
Jesse calls it “Superpowers” for good reason. This isn’t about adding new capabilities; it’s about making agents reliably execute on systematic, disciplined workflows. Every. Single. Time.
The Transformation: From Supervision to Systematic Development
The shift is dramatic. Here’s what actually changes when you use Superpowers:
TDD Becomes Automatic
The RED-GREEN-REFACTOR cycle happens without prompting. Claude writes a failing test first, implements just enough code to make it pass, then refactors. It doesn’t skip steps. It doesn’t take shortcuts. The discipline is built in.
Parallel Work Actually Works
Git worktrees get created automatically when you start new features. No more context confusion. No more “wait, which branch am I on?” Multiple Claude instances can work on different features simultaneously without stepping on each other.
Brainstorming Before Building
Instead of making assumptions and charging ahead, agents now ask questions. They explore alternatives. They validate designs in increments. The brainstorming skill ensures that coding doesn’t start until the problem is actually understood.
Subagents Follow the Same Discipline
When Claude dispatches subagents to implement tasks, those subagents use the same skills. The systematic approach cascades. You’re not just getting one disciplined agent—you’re getting an entire team of them.
The Meta-Breakthrough
Here’s where it gets wild: the skills test themselves using subagents. Jesse had Claude pressure-test whether future-Claude instances would actually follow the skills by simulating realistic scenarios with time pressure, sunk costs, and other factors that typically lead to cutting corners. Then Claude strengthened the skill instructions based on what made the test agents comply.
It’s TDD for process documentation. And it works.
My Personal Result
I mentioned this on the AIMUG Discord, and it still blows my mind: My personal output now exceeds what my entire teams at Oracle Cloud Infrastructure could produce.
This isn’t just about speed. It’s about systematic, disciplined development at scale. Subagents accelerated me, but Superpowers + the TDD workflow instantiates what I was trying to achieve through manual supervision. Instead of telling agents what to do, I’ve encoded the how into skills that compound into exponential productivity gains.
The breakthrough isn’t making agents smarter—it’s making them more disciplined and systematic.
Get Started Yesterday
Here’s how simple it is to install Superpowers:
/plugin marketplace add obra/superpowers-marketplace
/plugin install superpowers@superpowers-marketplaceTwo commands. That’s it.
Quit and restart Claude Code, and you’ll see the bootstrap prompt that kicks off the entire system. From that moment forward, Claude has access to skills that guide its behavior—from brainstorming to TDD to git workflows.
This is a community-driven repository. The skills can be shared, improved, and expanded by anyone. If you discover a better workflow or encode a new discipline, you can contribute it back.
This Is the Shift
What Jesse built with Superpowers represents a fundamental shift in how we work with AI coding assistants. We’re moving from ad-hoc supervision to systematic, repeatable workflows. From hoping our agents will follow best practices to ensuring they do.
If you’re using Claude Code, install Superpowers. Not tomorrow. Not next week. Yesterday.
This is the right definition for this repository. It gives your agents actual superpowers. And it gives you the ability to work at a scale that was impossible just a few months ago.
I encourage everyone to collaborate on it, contribute to it, and make it even better. This is what systematic development with AI looks like—and we’re just getting started.
Resources
- Jesse Vincent’s Original Blog Post - Deep dive into the thinking behind Superpowers
- Superpowers GitHub Repository - Install instructions and source code
- AIMUG Discord - Join the AI community where we discuss these breakthroughs