upvote
I remember before I started using AI (I was pretty late to the game), I assumed all the bugs and instability were because people were writing software with AI. Now that I've seen how good the code that AI writes can be (CAN be), I'm moving more to the thinking that it's from AI being used to find vulnerabilities.
reply
I've never had AI one-shot even a 10 line script that I didn't manually revise. To be fair, it usually produces something that works, but after seeing it I'll realize that my prompt was too vague or I left out some edge case. So it's the old problem with software "it does what was asked for but not what was needed."

So I speculate that a lot of problems with AI code are just that it is accepted uncritically because it builds and doesn't have any syntax errors.

reply
There’s a pressure to deliver more and faster from management. So instead of using LLM to reduce the coding time and allocate more time for review, everything is scaled down and there’s even less time for review. And you’re responsible for more as teams are also scaled down.

Not really a great recipe for good code.

reply
It’s more that management pushes more output and with AI it’s very easy to achieve but that will create slop.
reply
I think majority of companies don't know how to build safely with AI. It's just kind of handed off to engineering and mostly used as one-off chat interaction to get stuff done. Very few are building any kind of harnesses around SDLC that focus on architecture, testing, security and etc. I think we'll get there at some point, but in my experience the smaller tech players are just now on-boarding to use AI in general in any kind of capacity. We're not talking about best practices yet.
reply
Wondering if this has anything to do with the QA function being folded into the dev team in the last 15 years or so.

I always defended this, but it appears maybe we don’t have enough developers that take pride in the work they ship. LLM coding made this pretty obvious.

reply
It’s not lack of pride. It’s because they can’t say: “ok development is done, now I need a few days to thoroughly test this”.
reply
Engineers don't really decide what gets shipped at a typical tech company, you either ship trash when the marketing/MBA/'strategy' NPC says or they'll find someone else who will.
reply
> I also think AI can be used to build very high quality software.

I don't, and I think your comment is rhetorically useless because you've been bullied into making this ridiculous concession.

reply
bro... what model did you use to write that? or is that how you actually talk? Lol
reply
IMHO the only way to build good things is to pay attention to the details. And coding agents, while greatly increasing the appearance of productivity, they are causing people to pay less attention to what they're actually producing. This is the inevitable and obvious result.

Agentic coding could be a world-changing benefit, but the current way it's being sold and used as as a cost cutting measure -- produce software faster than ever before with fewer software engineers! Get rid of those pesky experts!

reply
We're in the v1 machines screws stage where everything is worse but with a clear path to get better .... I hope
reply
It isn't a software or tool problem.

It is a C-level executive problem. With the FAA it is the agency directorship problem. Same issue.

These people, very literally, are trying to run the world with power point presentations and executive summaries drafted by sycophants. 70 years ago these people had actual experience doing actual work and trying to grow businesses and doing stuff at the ground level. Now they are given their positions through a combination of credentials, connections, and political maneuvering. None of which actually relates to how well they can or cannot run a organization.

When reality disagrees with what they get in their emails and presentations then it reality that is wrong. It is the messengers that are the problem, not their competency.

This is nothing new. But the failures in leadership compound over time, like interest to a unpaid debt. They start off with healthy growing companies and thanks to government protections and bailouts it takes decades to for these organizations to decay into total dysfunction. In the past these sorts of people would be removed and their assets reallocated to competent management through bankruptcy and competition. However that natural process is not allowed anymore.

All AI is doing is handing them more foot-guns.

They think they can do better/faster/stronger/cheaper with AI and if AI causes problems then it can just be fixed by paying for better AI. Which they assume is just right around the corner, always.

Just wait until they start to realize that it is

reply
Very well put, and I think (open) AI will help competent people outcompete these clowns, which is what makes me optimistic. AI is a multiplier, but multiplying bullshit just creates more bullshit. These people are far more replaceable than competent engineers, even more so with AI.
reply
deleted
reply
broken windows
reply
Part of being sane is accepting that some tools, while being very powerful, are too easily misused to be widely distributed to those lacking proper training. This is why you cannot just go and buy 5 kilos of azithromycin
reply
Ehh I don't really buy that. Who decides what tools are distributed to who and how?

Don't want to get into a political debate here but I'll say that I think that open AI models and access to them is a benefit to society, while closed, proprietary models, controlled by tech billionaires are probably net bad for society.

reply
> Who decides what tools are distributed to who and how?

The same people we already delegate such responsibilities to: the government; ideally, a representative one.

reply
> the government

Uhhhh... you mean the corrupt, baby-eating pedophiles who are owned by tech billionaires right?

Cool plan bro

reply
> some tools, while being very powerful, are too easily misused to be widely distributed to those lacking proper training.

Could you define 'proper training'? I've endured plenty of fucked-up shit from so-called 'professionals/ experts/ trained people' over the last 30 years or so.

I'm currently assisting a couple of vibe-coders improve their 'too easily misused...[and] widely distributed' (available on their sites, or github) tools - I mean slop - that benefit me in ways traditional companies have no interest in. And the updates are fast (daily, mostly) and accurate because, unlike jeremy with a pretty CS degree and a nice job, these people actually give a crap' and don't have a manager to say 'you can't do that'. Also part of being sane is accepting that the cat is out of the bag and big $oftware/ tool manufacturer better get a move on. Disclosure: my interests are not critical to anything; quite the opposite, but people are learning faster and faster with Ai learning faster and faster. Who needs azithromycin if a series 1's and 0's can accomplish so much more? For good or bad.

reply
No, i cannot. If i could, i would be proposing concrete ideas. But I also understand why I cannot buy 3 kilos of plutonium or 5 kilos of antibiotics, and understand the modern ML models can do just as much damage while being more available.
reply