How I think about AI-assisted engineering
Most engineers use AI as a search engine with better formatting. You type a question, get a code block, copy it, adjust it. That is useful. It is not what I mean by AI-assisted engineering.
The shift
The shift happens when you stop asking AI to answer questions and start giving it context and a role. Instead of 'how do I do X', you say: here is the codebase context, here is the problem, here is the constraint, generate a solution that fits. The output quality difference is significant.
The question is not 'can AI write this code?' It is 'what is the right task decomposition so that AI can contribute at each step without introducing drift?'
What collaboration actually means
Collaboration means the AI is doing work I have defined, not work it has chosen. That requires explicit task scoping, structured output where possible, and a review step where I verify the output against intent. This is not that different from how you work with a junior engineer.
The honest limits
AI-assisted engineering is much worse at architecture than at implementation. It will give you a plausible architecture for almost anything, but plausible and correct are not the same thing. The decisions that matter, data model tradeoffs, API design, security boundaries, still require a human who understands the full context.
Contents
- The shift
- What collaboration actually means
- A concrete example
- The honest limits