Why agents do not write most of our code – A reality check
1–10 of 27 posts
Re: Why agents do not write most of our code – A reality check
#2Eerily maps to my experience almost word for word. I had codex write a chunk of code step by step with guidance and whatnot. Had to spend days cleaning up the mess.
Re: Why agents do not write most of our code – A reality check
#3great stuff; I've had almost exactly the same experience. I think blow-by-blow writeups like this are a sorely needed antidote to the hype
Re: Why agents do not write most of our code – A reality check
#4Eerily maps to my experience almost word for word. I had codex write a chunk of code step by step with guidance and whatnot. Had to spend days cleaning up the mess.
My experience is that if AI creates the mess, AI should clean it up, and it usually can, if you put it in a suitable agent loop that does a review, hands off small, well defined cleanup steps to an agent, and runs test suites.
If you review the first-stage output from the AI manually, you're wasting time.
You still need to review the final outputs, but reviewing the initial output is like demanding a developer hands over code they just barely got working and pointing out all of the issues to them without giving them a chance to clean it up first. It's not helpful to anyone unless your time costs the business less than the AI's time.
Re: Why agents do not write most of our code – A reality check
#5[dead]
Re: Why agents do not write most of our code – A reality check
#6One thing I find that constantly makes pain for users is assuming that any of these models are thinking, when in reality they're completing a sentence. This might seem like a nitpick at first, but it's a huge deal in reality: if you ask a language model to evaluate whether a solution is right, it's not evaluating the solution, it's giving you a statistically likely next sentence where yes and no are fairly common. If you tell it's wrong, the likely next sentence is something affirming it, but it doesen't really make a difference.
The only way to use a tool like this is to give a problem that fits context, evaluate the solution it chugs at you and re-roll it if it wasn't correct. Don't tell a language model to think because it can't and won't. It's a way less efficient way of re-rolling the solution
Re: Why agents do not write most of our code – A reality check
#7This has always been true. The difference is that now more people are admitting it. While you could argue that LLMs have junior level capabilities, they definitely do not have junior level self reflection or self awareness or self anything. It fundamentally doesn't learn where learning means being significantly less likely to fail at a task class x after being taught about it. And even just the ability of asking for help. These agents just choose to generate unusable code over stopping and asking for help or guidance and this implies that they are unable to tell their limits skill wise, knowledge wise, etc.
Frankly, I have been highly concerned seeing all the transformer hype in here when the gains people claims cannot be reliably replicated everywhere.
The financial incentives to make transformer tech work as it is being sold (even when it might not be cost effective) need to be paid close attention because to me, it looks a bit too much like blockchain or big data.
Re: Why agents do not write most of our code – A reality check
#8Eerily maps to my experience almost word for word. I had codex write a chunk of code step by step with guidance and whatnot. Had to spend days cleaning up the mess.
My experience is that if AI creates the mess, AI should clean it up, and it usually can , if you put it in a suitable agent loop that does a review, hands off small, well defined cleanup steps to an agent, and runs test suites. If you review the first-stage output from the AI manually, you're wasting time. You still need to review the final outputs , but reviewing the initial output is like demanding a developer hand…
IA reviewing code generated by AI, it's a recipe for disaster.
Re: Why agents do not write most of our code – A reality check
#9One thing I find that constantly makes pain for users is assuming that any of these models are thinking, when in reality they're completing a sentence. This might seem like a nitpick at first, but it's a huge deal in reality: if you ask a language model to evaluate whether a solution is right, it's not evaluating the solution, it's giving you a statistically likely next sentence where yes and no are fairly common. If…
but it’s also true that the next sentence is generated by evaluating the whole conversation including the proposed solution.
my mental model is that the llm learned to predict what another person would say just by looking at that solution.
so it’s really telling whether the solution is likely (likely!) to be right or wrong
Re: Why agents do not write most of our code – A reality check
#10Earlier quoted context omitted.
My experience is that if AI creates the mess, AI should clean it up, and it usually can , if you put it in a suitable agent loop that does a review, hands off small, well defined cleanup steps to an agent, and runs test suites. If you review the first-stage output from the AI manually, you're wasting time. You still need to review the final outputs , but reviewing the initial output is like demanding a developer hand…
IA reviewing code generated by AI, it's a recipe for disaster.
That's categorically not true, as long as there's a human reviewer at the end of the chain. It can usually continue to deliver actual improvements over several iterations (just like a human would).
That does not mean you can get away with not reviewing it. But you can most certainly with substantial benefit defer reviewing it until an AI review thinks the code doesn't need further refinement. It probably still does need refinement despite the AI's say so - and sometimes it needs throwing away -, but it's also highly likely in my experience to need less, and take less time to review.