Code Like a Surgeon: Maximizing Developer Productivity with AI

In the world of software development, the narrative around AI often positions it as a tool that will turn us all into managers or editors. However, this perspective is dangerously incomplete. Instead, I’m adopting a different mindset: coding like a surgeon.

What It Means to Code Like a Surgeon

A surgeon isn’t just a manager; they are deeply involved in the actual work. They leverage a support team for preparatory tasks, secondary responsibilities, and administrative duties, allowing them to focus on the critical aspects where their expertise is indispensable. Similarly, my goal with AI coding tools is to spend 100% of my time on high-impact tasks. For me, as a UI prototyper, this means tinkering with design concepts.

Leveraging AI for Secondary Tasks

There are numerous secondary tasks that AI agents can handle effectively:

• Codebase Guides: Before tackling a major task, I have an AI write a guide to the relevant areas of the codebase.
• Spiking Changes: I often ask an AI to spike out a significant change. Even if I don’t use the result, it provides a useful sketch of where to go.
• Bug Fixes: For clear-cut issues in TypeScript, I delegate bug fixing to AI agents.
• Documentation: Writing documentation about what I’m building is another task that can be offloaded.

Running these tasks asynchronously in the background-while eating lunch or even overnight-ensures that everything is ready when I sit down to work. This approach mirrors a surgeon walking into a prepped operating room, where all the preparatory work has been done.

The Autonomy Slider

There’s a significant difference between how I use AI for primary versus secondary tasks:

• Primary Tasks (Core Work): For core design prototyping, I still do a lot of coding by hand. When I use AI, it’s with careful attention to detail and fast feedback loops. Tools like Cursor tab-complete are invaluable here.
• Secondary Tasks: For these, I’m much more relaxed, allowing agents to run for hours in the background. The priority is getting the job done, even if it takes longer. Claude Code has been my go-to for long unsupervised sessions, but Codex CLI is becoming a strong contender and might become my new favorite.

These distinct work patterns align with Andrej Karpathy’s concept of the “autonomy slider.” It’s crucial not to conflate different parts of the autonomy spectrum because the tools and mindset required vary significantly.

Your Agent Doesn’t Need a Career Trajectory

The idea of a "software surgeon" is not new. Fred Brooks attributes it to Harlan Mills in his 1975 classic, “The Mythical Man-Month.” Mills proposed that each segment of a large job should be tackled by a team, but the critical work should be done by a single person-the software surgeon-supported by a team for preparatory tasks.

This concept is more relevant than ever in the age of AI. Just as a surgeon’s support team doesn’t need to worry about career trajectories, your AI agents don’t either. They are there to handle the secondary tasks efficiently, allowing you to focus on what truly matters.

Conclusion

By coding like a surgeon, you can maximize your productivity and leverage AI tools effectively. This approach ensures that you spend your time on high-impact tasks while offloading secondary responsibilities to AI agents. The key is to understand the autonomy slider and use the right tools for each type of task.