"It's simple. All we need to do is agree on exactly what we mean by 'bad code' and then codify all of that - including all the tradeoffs and value judgements informed by decades of experience but that we've never articulated and that are highly context-dependent - in a computable form that runs really fast"

But even if you *could* make quality gates good enough that you won't need to review the code - and good luck with that - you then run straight into another trap: comprehension debt.

LLMs will never be reliable enough that you won't need to understand the code. That's physics.

@jasongorman I feel like outsourcing comprehension is not a wise move. Yet people with 20+ years of experience think it's fine, because "you can just rewrite from scratch anytime."

@jasongorman also: the context can change, and recognising that requires contextual people skills.

By which, I mean that: you have a context - you're told its a throwaway poc, say. Do you trust the person telling you that? Do you trust the organisation not to change its mind later? All of these are meta contextual. Good luck teaching your LLM that.

@rogerlipscombe Software development will always require Actual Intelligence being applied throughout. There is no "feed it a spec and go to the pub" - probably not in our lifetimes. Everyone's chasing a fantasy, IMO.

This is a good plan. We can just write The Code and then we will have finished The Computer.

@jasongorman