Most developers ask AI tools to write a function or fix a bug. That’s fine, but here’s the cheat code I discovered: treat AI like a developer on your team. When I do that, everything changes.
The Pattern I Discovered
I use AI tools every day in my professional work and personal projects. Claude, ChatGPT, Gemini; I subscribe to all of them. Over time, I’ve noticed a pattern.
These tools excel at zooming in. By that, I mean they’re excellent at specific tasks: generating code, writing tests, refactoring functions, etc… . We all know this.
However, they struggle with connecting the dots across systems with many moving parts. That responsibility falls on us.
When I treat AI like an implementer, giving commands like “do this, then do that”, I get decent results. But when I step into a leadership role and set the architecture, clarify constraints, and define interfaces, I get dramatically better results.
My prompts become more effective, not because I memorize every file or function in the codebase, but because I understand the system end-to-end. I connect the dots and let AI handle the implementation details.
Here’s how I approach it: I zoom out before I zoom in. First, I examine the architecture and identify the outcomes I’m targeting. Then I zoom in with AI coding tools like Claude or Cursor IDE to execute the specific components I need.
I don’t remember everything, I just need to understand where each piece fits and why. These tools help me move quickly within modules, while I remain responsible for connecting the dots between modules.
Why This Matters Now
Currently, especially in systems with multiple services, queues, databases, and APIs, coding tools cannot connect everything for you. Perhaps they will in the future, but today, the crucial skill is being able to hold the architecture in your head and guide the tools to implement the details.
This skill multiplies the value you’ll extract from AI.
Why is this important right now? Because the ability to connect systems is both scarce and valuable. I’ll be honest, I believe it was always scarce.
But we’re entering an era where this skill will be more critical than ever. AI tools are already strong at local code generation, but they lack reliable understanding of cross-service context and cannot make system-wide trade-offs because they don’t have the complete picture.
If you can provide technical leadership, AI will help you build remarkable things.
Watch the Video
I also shared this AI development approach in video format. You can watch it here: