In our most recent podcast interview with Google's Principal Engineer Taylor Mullen, Taylor mentioned something that deserves its own spotlight: Test-Driven Development (TDD). If you're using AI to write code, this one concept could be the difference between "it works" and "it works and I trust it."
Here's the simple version: TDD means you write a test before you write any code. The test describes what the code should do. Then you let the AI build only enough code to pass that test. It's a three-step loop developers call Red, Green, Refactor:
- Red: Write a test. Run it. It fails (because the code doesn't exist yet).
- Green: Write the minimum code to make the test pass.
- Refactor: Clean up the code while keeping all tests green.
Why does this matter for AI? Because telling an AI "build me a login page" is vague. But handing it a set of tests that define exactly how that login page should behave (what happens when the password is wrong, what the button looks like while loading, etc.) gives the AI a concrete target. As developer David Luhr put it: TDD is prompt engineering. The tests tell the AI both "what" to build and "when it's done."
Think of it like giving a contractor a blueprint instead of saying "build me something nice."
The tool everyone's talking about: Superpowers
Jesse Vincent, a veteran open-source developer (he created Request Tracker and co-founded keyboard company Keyboardio), built a plugin called Superpowers that bakes TDD directly into your AI coding workflow. It started as his personal system for getting better results from Claude Code, and it took off. 42,000+ GitHub stars. Simon Willison, one of the most respected voices in AI tooling, called Jesse "one of the most creative users of coding agents that I know." Anthropic officially added it to their marketplace in January 2026.
What Superpowers actually does:
It enforces a structured workflow that keeps your AI agent on the rails. Think of it as guardrails for your robot coder:
- Brainstorming: Before writing a single line of code, the AI asks you clarifying questions about what you're building. No more "I assumed you wanted purple buttons."
- Planning: It breaks your project into bite-sized tasks with specific success criteria, so the agent knows exactly what "done" looks like.
- TDD implementation: For every task, the agent writes a failing test first, builds only enough code to pass it, then refactors. If it tries to skip the test? The skill literally tells it to delete the code and start over.
- Code review: After each task, a separate AI "reviewer" checks the work against the original plan. Did it actually build what you asked for, or did it wander off?
- Sub-agent orchestration: It can dispatch multiple AI agents to work on different tasks simultaneously, each following the same disciplined process.
How to get started (even if you've never used a terminal)
Developer Rob Shocks made an excellent walkthrough video building a real app with Superpowers from scratch. A few timestamps worth bookmarking:
- (0:54) The Superpowers workflow explained in 60 seconds
- (5:13) Step-by-step brainstorming with the AI
- (10:01) The agent working autonomously for 40 minutes while Rob gets coffee
- (13:36) Superpowers vs. vanilla plan mode comparison; the results might surprise you
If you already use Claude Code, setup is two commands:
/plugin marketplace add obra/superpowers-marketplace /plugin install superpowers@superpowers-marketplace
Restart Claude Code. That's it. The skills activate automatically; you don't need to memorize commands.
It also works with Cursor, Codex, and OpenCode (setup instructions on the GitHub repo).
P.S: We also published a full guide to the top AI plugins for coding agents; check it out if you want more plug-ins like this one!
