upvote

points

by shsush1 days ago |
comments
upvoteby submain1 days ago|
next
[-]
> The real secret to agent productivity is letting go of your understanding of the code and trusting the AI to generate the proper thing

The few times I've done that, the agent eventually faced a problem/bug it couldn't solve and I had to go and read the entire codebase myself.

Then, found several subtle bugs (like writing private keys to disk even when that was an explicit instruction not to). Eventually ended up refactoring most of it.

It does have value on coming up with boilerplate code that I then tweak.

reply
upvoteby maplethorpe1 days ago|
parent|
[-]
You made the mistake of looking at the code, though. If you didn't look at the code, you wouldn't have known those bugs existed.
reply
upvoteby PunchyHamster1 days ago|
parent|
[-]
fixing code now is orders of magnitude cheaper than fixing it in month or two when it hits production.

which might be fine if you're doing proof of concept or low risk code, but it can also bite you hard when there is a bug actively bleeding money and not a single person or AI agent in the house that knows how anything work

reply
upvoteby urig1 days ago|
prev|
next
[-]
That's just irresponsible advice. There is so little actual evidence of this technology being able to produce high quality maintainable code that asking us to trust it blindly is borderline snake-oil peddling.
reply
upvoteby hansmayer1 days ago|
parent|
[-]
Not borderline - it is just straight snake-oil peddling.
reply
upvoteby _zoltan_1 days ago|
parent|
[-]
yet it works? where have you been for the last 2 years?

calling this snake oil is like when the horse carriage riders were against cars.

reply
upvoteby hansmayer1 days ago|
parent|
[-]
I am an early adopter since 2021 buddy. "It works" for trivial use-cases, for anything more complex it is utter crap.
reply
upvoteby Kubuxu1 days ago|
prev|
next
[-]
I don’t see how I would feel comfortable pushing the current output of LLMs into high-stakes production (think SLAs, SRE).

Understanding of the code in these situation is more important than the code/feature existing.

reply
upvoteby shsush1 days ago|
parent|
next
[-]
I agree and am the same. Using them to enhance my knowledge and as well as autocomplete on steroids is the sweet spot. Much easier to review code if im “writing” it line by line.

I think the reality is a lot of code out there doesn’t need to be good, so many people benefit from agents etc.

reply
upvoteby danenania1 days ago|
parent|
prev|
[-]
You can use an agent while still understanding the code it generates in detail. In high stakes areas, I go through it line by line and symbol by symbol. And I rarely accept the first attempt. It’s not very different from continually refining your own code until it meets the bar for robustness.

Agents make mistakes which need to be corrected, but they also point out edge cases you haven’t thought of.

reply
upvoteby Kubuxu1 days ago|
parent|
[-]
Definitely agreed, that is what I do as well. At that point you have good understanding of that code, which is in contrast to what the post I responded suggests.
reply
upvoteby heavyset_go1 days ago|
prev|
next
[-]
> The real secret to agent productivity is letting go of your understanding of the code

This is negligence, it's your job to understand the system you're building.

reply
upvoteby hansmayer1 days ago|
prev|
next
[-]
Not to blow your bubble, but I've seen agents expose Stripe credentials by hardcoding them as text into a react frontend app, so, no kids, do not "let go" of code understanding, lest you want to appear as the next story along the lines of "AI dropped my production database".
reply
upvoteby yonaguska1 days ago|
prev|
[-]
This is sarcasm right?
reply
upvoteby PunchyHamster1 days ago|
parent|
next
[-]
I wish, that's dev brain on AI sadly.

We've been unfucking architecture done like that for a month after the dev that had hallucination session with their AI left.

reply
upvoteby shsush1 days ago|
parent|
prev|
[-]
[dead]
reply