Eating My Words

It’s been about two years since I wrote a blog post about my opinion on AI and code generation, and I think it’s about time I’ve revisited my perspective. In short, I’d say I was completely wrong about my skepticism in integrating it into my day to day coding. After slowing letting AI into my workflow, there is a real benefit in terms of productivity and understanding.

I recently started moving from a product engineering focus role and into a more of a data platform role. I think, without AI, I wouldn’t have been able to bring myself up to speed with all the new things I had to learn. And I think that’s the key, is to let AI help you learn what you don’t know. It’s easy enough to just write a prompt and tell AI what you want and hope for the best. But, I found that during this transitions, my prompts have been more exploratory in nature. Taking on the role from a point of ignorance help AI give me the framework I needed to be more comfortable and take on greater challenges.

To be fair, I still do believe we shouldn’t be complacent in just letting AI generate mode code than we actually need. We should still look into abstractions and frameworks that compress knowledge into digestable chunks. After all, AI context is a finite resource at the moment and reducing the surface level of what you AI can write will also help in reducing the amount of bugs it could possibly introduce.

If you haven’t started using any of the frontier models and agents already, I suggest you do. Take it slow at first. Treat it as a sharp tool that helps you disect what you are doing. Work on prompts that guide you and your agent into uncovering the work. Don’t treat prompts like a hammer and hope for the best. AI should be your chisel that helps you craft with percision.

PS: This post was NOT written by AI.