In my career I've been a six nines systems engineer. Stuff like billion dollar daily volume payments platforms, and my work was always situated in critical flows - so I know a thing or two about solid engineering work.
Just a moment ago I asked AI for a complicated multithreaded state machine with variational queuing logic. Claude Code Opus delivered. I'm reviewing it now. It's phenomenal code.
You need to reassess the world. Your priors are deeply flawed. These models are absolutely incredible.
I mean it - you're going to be hurt badly if you don't reevaluate what is happening and plan accordingly.
I think you'd be the first one, so I highly doubt that. If it's true, good on you.
I use Opus daily. It can take some typing off my hands, as long as I keep it to highly specific, limited, straight-forward things. And as long as I spend a long time preparing everything in the most minute detail. Then I have a chance of it being a slight efficiency gain. Veer slightly outside of those preconditions and the output is invariably impressive at first glance, garbage at second glance. Not to mention that it's barely any help for the 80-90% of the job that isn't writing code.
Even Jensen Huang expects the LLM efficiency boost to be about 30% for software engineers. Jensen freaking Huang, who has every reason to exaggerate the benefits! So I'm realistically taking that as an upper bound.
Run rate is easy to get while not meaning much (which is why GenAI vendors love talking about it). Report back in a year about how things are going, and how much of your code you had to rewrite from scratch.
What plans can most people reasonably come up with?
If AI is taking over software, it's taking over lots of other computer related jobs. If you're not a highly paid engineer, or come from money, most people can't just re-skill for entirely different careers.
Point taken. I'll just ask the AI to make me a really fun AAA multiplayer game next time I work on one.
