Most people using AI to code are still writing code — just faster. They autocomplete lines, generate functions, ask for refactors. The AI is a tool in the loop.
I've been working differently. I don't write the code. I set direction, define constraints, and decide what ships. The agents handle implementation — writing, testing, debugging, iterating.
This isn't delegation in the traditional sense. It's closer to creative direction. You need to think clearly about what you want, communicate it precisely, and know when the output is right. The skill shifts from typing to judgment.
The mental model
Think of it like running a studio. You're the creative director. The agents are the team. You don't micromanage the brushstrokes — you set the vision, review the work, and course-correct.
The workflow looks like this:
- Define the intent. What are we building? What constraints matter? What does "done" look like?
- Let agents execute. They write code, run tests, fix errors. Sometimes they get stuck and you redirect.
- Review and ship. You're the quality gate. Does this match the intent? Is it clean? Does it work?
What actually changes
The biggest shift is psychological. You stop identifying as a "coder" and start identifying as a "builder." The implementation is a means to an end — the product is what matters.
You also get faster at the parts that matter: architecture decisions, design direction, system thinking. The parts that were always the bottleneck.
What's next
I'm building an orchestration layer — multiple specialized agents that handle different parts of the workflow. Planning agents, execution agents, review agents. Each one optimized for its role.
This website is built that way. The activity feed you see on the homepage is real — it shows the agents working in real time.
More on that soon.