AI Coding Agents Need Fundamentals

I’ve been spending more time with AI coding tools like Cursor — and at first, I was struggling to get quality results.

They worked great for code inspection and small updates, but fell short for feature development and large refactors.

At a local AI conference, Chuck Russell shared his approach and it reminded me of something basic: the fundamental software engineering process still matters.

I would never ask a team to “just start coding” without thinking about users and planning the work. AI agents are no different.

Discovery – Use ChatGPT to process Fathom / Gemini / Zoom transcripts and pull out themes, pain points, and opportunities.

Requirements – Turn those insights into clear requirements and specs for new features.

Planning – Give the coding agent the spec and ask it to propose an implementation plan.

Then you’re ready for the agent to build the feature.

AI coding agents have come a long way. They accelerate the work, but they don’t replace the fundamentals. You still need an experienced human to guide the process.

What are you seeing? Have your teams successfully integrated AI coding agents into your workflows?

Originally published on LinkedIn — view the original post for comments and reactions.