I started a new job about two months ago, and the team uses AI for a large part of the programming work. My previous job used very little AI, so this has been a major adjustment. I can prompt tools such as Copilot in plan mode and often get working code, but I worry that I am completing tasks without actually building my programming skills.
What is the best way to learn the codebase, become useful, and continue progressing in an AI-heavy development environment? If you use AI for much of your work, what do you do while it is generating a solution, especially when that can take anywhere from a few minutes to several hours? I would appreciate advice from people who have adapted to this kind of workflow without letting the AI replace their understanding.
4 Answers
AI may make experienced engineers much faster, but experience still matters because it provides the foundation for judging what the tool produces. Do not optimize every task purely for speed. Pick a skill you want to improve, spend time investigating the problem yourself, try a few approaches, and use AI to compare ideas, explain unfamiliar concepts, generate tests, or point out edge cases.
AI can help you get something working, but make yourself explain the result before accepting it. Try rebuilding small sections without assistance, trace the code with a debugger, and write down why the solution works. That turns AI from an answer machine into a learning tool.
Your biggest advantage is understanding the project and its architecture. Spend time learning how the systems fit together, where data enters and leaves, what the important constraints are, and how the team tests and deploys changes. The better your engineering context, the better you can guide the AI and catch its mistakes.
Focus on the parts relevant to your current work rather than trying to understand everything. Read recent changes, follow a feature from its entry point through the system, and ask the AI to explain the architecture only after you have formed your own understanding.
When the tool is working, do not just wait passively. Read the surrounding code, inspect documentation, review the plan line by line, predict what the implementation should do, and prepare tests. You can also take one small part of the task and solve it manually before comparing your approach with the generated code. Staying actively involved is what prevents the workflow from becoming pure prompting.
That is difficult when the codebase seems to change every time I look away. How do you keep up?