Observations Around Prominent Programmer Sentiment and AI

A few months ago, it was easy to find AI sceptics among prominent programmers. Just over a year ago DHH was interviewed on the Lex Friedman Podcast. Now appearing again on the recent episode, DHH's opinion seems to have shifted significantly.

I find there are more and examples of this sentiment out there. While Andrej Karpathy (Former Director of AI at Tesla, OpenAI Co-founder) has always been ahead of the curve on his bullishness regarding moving to AI workflows (possibly due to his proximity to the engineers and labs working on the latest models and tooling and his deep understanding of the field), we are seeing more engineers 'admit' to spending far less time writing code by hand or reviewing the code that AI generates - rather using phrasing like 'reviewing the shape of the code'. Clearly there is greater trust in agentic coding outputs. DHH made the analogy that people used to complain that satellite navigation systems would direct people into water, but that nobody thinks that this happens anything like frequently now. AI is not perfect, but it is on the same trajectory.

Personally, I can believe this is true, but with caveats. The smaller the task, the truer this is. Letting agents run on large complex tasks still results in software that does not match my taste in terms of behaviour, performance and sometimes security. More and more often, it is debatable whether these 'defects' are matters of opinion, but it still highlights the pitfalls of ceding control, and reduced understanding of a code base.

I agree with some that even when we do fully understand a codebase at the time of writing, it is easy to forget if we are not always working on the same project. I also believe that we are now at the point where agents are generating better code than your average development team used to, when you factor in that significant amounts of development work that was undertaken by junior and mid-level engineers.

I still feel there is a huge bottleneck around attention, and I think this is going to hurt us in terms of technical debt and security issues. I'm fairly certain that we will attempt to solve both problems with more code, and I'm unsure where this ends. For the time being, I will observe with interest and do my best to keep up with this fast changing landscape.