Post

I’ve gotten this a couple times so one thing to clear up: At this point I think AI coding is akin to having a new compiler, and the language is just natural text (often specs). LLMs do not currently have the capability to make good engineering decisions past a certain level of complexity

I’d also wager that getting AI to do good work in any profession requires the same level of domain expertise to check and guide the work, eg law

I’ve seen people who don’t have engineering backgrounds be able to prompt their way to creating an output that looks like what they think they want. The problem is that if you look under the hood, it’s not well-architected, and there’s lots of issues with the code.

I’m worried that what may happen is that execs start seeing non-technical employees start turning these things out, think it’s good enough because they don’t know the right questions to ask, and then start questioning why they employ expensive engineers who take longer