If You Can’t Explain It, Don’t Ship It

As a software developer, writing code is only a small part of the job — the real work is in thinking, making decisions, and understanding systems deeply. That’s why the rise of AI in software engineering is both exciting and concerning. Tools like GitHub Copilot and Cursor have made code generation effortless, and while they’re useful, if I’m recommending one for serious workflows, it’s Claude — especially for terminal-based tasks where context and precision matter. But regardless of the tool, the responsibility still sits with the engineer.

Using AI blindly is where the problem begins. If a piece of code makes it into your system and you can’t explain it, that’s a red flag. If AI introduces a subtle bug or a security issue and it slips through because you didn’t review it critically, that’s not on the tool — that’s on you. In a real team, someone will ask why a decision was made or what a line of code is doing. “AI wrote it” is not an answer. As engineers, we are accountable for every line that ships.

The discipline is simple: you decide, AI executes. Before you ever prompt AI, you should already have clarity — almost like you’ve written the pseudo-code in your head. You understand the trade-offs, the edge cases, and how the change fits into the system. Then you can direct AI to implement your solution, not invent one for you. Review it, test it, and refine it until it meets your standard. The good part is that, you can use the same AI to research and come up with a clear thought process. This should always come first.

I’m curious — how are you using AI in your workflow today? What tools have actually improved your thinking, not just your speed? Have you found yourself relying on it too much, or has it made you a better engineer? Let’s hear your experience.