I had an odd thought recently. My goal, as a consultant, is to get myself replaced by AI.
Let me explain,
All AI technology is based on pattern recognition, and the newer models are getting really good at picking up on deep and subtle patterns that are not at all obvious.
So when engineers use AI code generation — for example GitHub Copilot, or Cursor, or many others) in a project, the underlying model is exceptionally keyed in to existing patterns in the code. If your project has, for example, similar bits of code copy-pasted all over the place, you’ll get completions that do more of the same, entrenching this pattern more and more deeply. (And this particular pattern is already obnoxiously difficult to root out)
If you have a clean architecture, with strong separation of concerns, limited coupling between modules, across-the-board tests that make it easy to refactor, and a focus on human-readability, AI will pick up on that too, and crank out more of the same.
Either way, the seeds of your project’s success or failure are already there.
Working on real-world projects, I’m gradually realizing that this is the key factor in modern projects distinguishing between those that succeed vs those that go over-budget, over-schedule, or even collapse under their own weight.
By the way: If you think that nobody on your team is using AI tools, you’re probably wrong, especially when tools are widely available and come natively integrated in common IDEs. So far, studies about code quality with AI code generation haven’t taken this into account, but I’ll bet this is responsible for some of the documented quality issues.
Managers and stakeholders: is your project’s core architecture AI ready? Contact me to find out.
Software engineers: keep up to date on the latest AI technologies with free weekly email updates — even if you don’t use them, it’s something you need to know about.