Post

Simon Willison
@simonwillison.net

At the risk of starting the flame war to end all flame wars... Modern LLMs (GPT-5.1, Claude 4.5, Gemini 3) produce excellent code and can be a significant productivity boost to software engineers who take the time to learn how to effectively apply them - especially if used with coding agent tools

Josh Vlk
@vlkpack.com

They can be useful for developers who are experts and know what the output should be. They are a hindrance to junior developers and you shouldn't use them if you are learning or you don't know how to evaluate the output.

Simon Willison
@simonwillison.net

That's still a big open question for me I've coached a lot of early stage developers and one of the key things I've found is that the first six months are agonizingly frustrating - you miss a semicolon and your code doesn't compile and it takes an hour to figure out why

1 repost28 likes

Jonny Buchanan
@jbscript.dev

This can also apply to developers tackling tech they're not familiar with, in complex apps, with LLM usage robbing them of what should have been a chance to learn. This also affects code review as a chance to learn, what do you do with critique of complex code you never understood? Plug it back in

Julius Volz
@juliusv.com

Yeah, I think the most important thing for junior developers to understand is that they shouldn't just trust LLM output and always still try to understand what's going on. But it's great for getting pointers in the right directions, discussing different architecture/implementation options, etc.