After 30 years in software, I learned that working with AI isn’t about the tool – it’s about developing the right skills. Clear prompts, breaking problems into chunks, careful delegation, and treating AI like you manage people makes all the difference.
The story
After three decades in software and nearly a decade building AI solutions, I’ve realized something: delegating to AI teammates isn’t about the tool. It’s about developing the skills to work with them effectively.
Here’s what actually matters:
Prompt engineering is everything. Vague instructions get vague results. When I need a sorting function, I don’t say “sort this list.” I specify: “Write a Python function that sorts integers using quicksort, handles duplicates, and returns a new list without modifying the original.” The more context and clarity I provide, the closer the output comes to what I’d expect from a human teammate.
Break problems into chunks. I treat AI agents like junior engineers – they thrive on well-defined, modular tasks. Instead of throwing a big problem at them, I break it down: write the data model, generate the parsing function, draft unit tests. This makes debugging and iteration way easier. I keep context files so the agent picks up where we left off between sessions.
Choose what to delegate carefully. I use AI for boilerplate, test generation, documentation, and initial drafts of non-critical features. Anything security-sensitive or core to the business gets human oversight. Security is always priority zero – AI-generated code needs the same review as code from any new hire.
Manage AI like you manage people. Give clear instructions, review their work, provide feedback. I never fully automate pull requests without review, and I document key learnings for each session. Over time, this builds a cheat sheet that makes future collaboration smoother.
Recent Comments