Now let’s go live to Amazon for the latest updates about this developing story.
Amazon’s ecommerce business has summoned a large group of engineers to a meeting on Tuesday for a “deep dive” into a spate of outages, including incidents tied to the use of AI coding tools.
The online retail giant said there had been a “trend of incidents” in recent months, characterized by a “high blast radius” and “Gen-AI assisted changes” among other factors, according to a briefing note for the meeting seen by the FT.
Under “contributing factors” the note included “novel GenAI usage for which best practices and safeguards are not yet fully established.”
↫ Rafe Rosner-Uddin at Ars Technica
Now where could all the regular AI boosters in the comments have disappeared to, I wonder?
They’re probably off restoring from backups after clawbot had a field day.
adkilla
Are we really dead set on this AI civil war enveloping every topic?
I for one am not against covering the downsides of AI, it’s important even. But don’t you honestly get tired of the same topic getting this much attention every week? There’s only so much to say about it before it all gets so repetitive.
AI is such a big field, let’s at least cover new ground. For example training neural nets to solve problems in new ways.
“It Took Me 30 Years to Solve this VFX Problem”
https://www.youtube.com/watch?v=3Ploi723hg4
I want topics to be diverse, fun, open minded, thought provoking, etc… I would think that everyone could benefit from taking the temperature down a little bit on our AI civil war, which I don’t think anyone is having fun with. I could be wrong and maybe you really do want to keep it going, but I prefer fresher topics.
Alfman,
Although I appreciate the sentiment behind what you’re saying, ignoring the debate isn’t a neutral action. As long as the tech industry keeps trying to sell AI as a cureall while sweeping all its limitations and disadvantages under the carpet, there very much is value in offering dissenting views, regardless of how repetitive they might be.
We’re being bombarded with pro-AI propaganda and society is being restructured for the sake of enriching the few while the world burns, and I’m deeply grateful to OSNews for continually being a voice of reason.
Aankhen
The thing is, we’ve gone well beyond the line of “ignoring the debate and neutrally”. That’s not in question and there is zero risk of that happening. We can all continue making the same points over and over again, but it’s just flogging a dead horse at this point. Come back to it now and again sure, but lately it feels like AI has become the main topic for osnews constantly. There are so many topics that are fun to learn and talk about and I find one of Thom’s greatest strengths is finding those diamond in the rough articles and projects that inform, educate, and entertain us. That’s fun and why I am here.
Seconded, signed, nodded and sealed.
My jab was specifically on coding and automation tools built on top of LLMs. Not on all ML initiatives.
My comment was on topic and is not an attack on AI. Unlike what the AI bros have been marketing, I don’t buy in on their hype that LLMs are all encompassing as ‘AI’.
I agree there’s so much more happening in AI. I’d prefer we covered many different kinds of AI here on osnews. My link was meant as an example of this.
Wow, you should really ask them to pay rent, in your mind.
(do not take the bait, do not take the bait)…
It seems like we will have a real rough ride while many organizations will use the tools in a wrong manner. We will probably look back to these days like the early internet or “excel” boom and see how silly some of these mistakes were.
Just for context, other companies have been using these or similar tools without issues. Just because they are not actively talking about it, does not mean they do not exist or beneficial.
The problem with “AI” is that it spits out a crapload of generated information of dubious correctness (and with no subtle verbal or other clues as to what could be correct and what could be wrong), so you have to painstakingly check the output, at which point you could just do it yourself. And yet, PHBs think they can fire half their employees because of the non-existent productivity boosts of “AI”.
kurkosdr,
I’ve just had dinner with some friends from Amazon, they are perfectly fine. The bottom line is, they learned a few lessons, and will continue to use these tools.
What I really fear is that this is going to increase the divide in several ways
\\
1 – The obvious one: We are reducing the opportunities for entry level jobs. Something an intern can do can now mostly be automated
2 – As always “haves” and “have nots”. A proper single session with Claude Opus can easily cost over $100. For an established company, or a rich individual, paying $100+ per PR, and paying $3,000 per month per employee is an absorbable cost. For a small company that is trying to establish themselves… more challenging.
\\
That is why we need better tools, democratized access, and more importantly better education on how to access these.
You can’t teach an ‘AI’ to become a senor dev. But you can mentor junior devs to become senior devs. Eventually, these corpos will move more of the work overseas in the hopes that underpaid and under trained devs in 3rd world countries can pick up the slack. The models have not been improving much and they aren’t getting cheaper. It is a race to the bottom until it bottoms out.
adkilla,
Yes, that is my point. The pipeline for junior -> senior developer progression is being cut.
(And let’s be honest we all did terrible coding when we were young. “Why do we need a for, I can add 10 print statements!”)
Not all of them does it. At least the companies who actually worry about their bottom line, and has two functioning brain cells.
That is demonstrably incorrect. I would suggest spending some time with modern ones like Gemini 3.1 Pro, or Claude Opus 4.5/4.6.
But not “I tried something and it did not work”. More of “I have spend a week and was able to build an entirely new automated log processing system from scratch”
(you need to be at senior level to actually get benefit out of those)
For example
https://www-cs-faculty.stanford.edu/~knuth/papers/claude-cycles.pdf
Here is Don. Knuth admiring Claude:
This is fascinating, since Knuth has literally written “the books” on software development. And he seems to be immensely enjoying the experience.
https://lwn.net/Articles/1061326/
Ted Ts’o
Andreas Reichel,
True.
And if we are pedantic… any computer system that does decision is AI. So, autocorrect, intellisense, grammar checks… all fall under this umbrella.
Between those who would swear by all the garbage spewed from naive use of AI, to those who want to ban even the mention of it…
Walking the tight path of balance is becoming harder.
sukru,
What are we referring to here? As a software engineering or CS student in school? I am sure they would have taught you the basics of imperative programming at school and have equipped you with the basics for a job.
You mean like Amazon? Hate to break it to you, they don’t: https://www.businessinsider.com/amazons-just-walk-out-actually-1-000-people-in-india-2024-4
I tried building a reporting tool in C# for Defender for Cloud consumption using Claude and GitHub CoPilot. Absolute trash. It hallucinated APIs that did not exist, it attempted to generate suppression for warnings, when I prompted it not to do that earlier on. I saved more time just writing it from scratch.
A job that would have taken me a day to complete it, took a frustrating 3 days with these tools. The time I wasted in this scenario, could have been better used to train a junior dev and they would have even been able to maintain the code in future. There would have been no need to repeat this frustrating cycle for the next endeavor.
I guess if you are writing a tool in Python or React, where there is a ton of example code and boilerplate out there, it might work. So much for replacing devs as what the AI bros keep telling us.
adkilla,
I was actually referring to middle school, but anyway…
(You really see those level of code in professional life, though. I you have not, you have been very lucky)
Obviously, not referring to them.
I think you made my point here.
The tools don’t make you faster or slower. But they make your more efficient. What does it mean? You can still get the same amount of PRs in the same amount of time. But they will have more breadth.
(Things like better testing, stronger adherence to coding standards, better evaluation of alternate solutions and edge cases, and so on)
sukru,
How does it help with anything you have claimed here? It can’t even live up to the marketing hype of making coding obsolete. Not everyone has a week of time to waste working on some boilerplate logging system. As I have stated: “I guess if you are writing a tool in Python or React, where there is a ton of example code and boilerplate out there, it might work.”
Claiming that other’s needs are insignificant, just because it does not align with your narrow definitions, is just lame IMHO.
Despite it not being able to solve the simple problem I gave it, I am supposed to hand these tasks to it too? Should I then waste more time correcting it’s mistakes and hallucinations for these too? I would like your opinion on this interview with Jeremy Howard: https://www.youtube.com/watch?v=dHBEQ-Ryo24
I’d be interest to see how balanced and honest you are.
adkilla,
I’m not sure why we are going in circles.
Anecdotes don’t matter as much, but results do.
This is the “summary” of that YouTube video you shared. And I don’t disagree with it.
And AI assistant that generates code will not magically make us better coders. If an engineer is unable to understand the code that is produced, it will make him even worse.
It is a force multiplier.
In both directions.
Let me tell it this way.
If one says
He would most likely get slop.
If one says
Would get a much better result.
sukru,
It’s because you make comments like these:
You mean results that don’t ‘align’ with your expectations? You aren’t convincing anybody.
He said nothing of the sort. Where did you get this made up “summary’? Did a LLM fabricate it for your? Here is the actual summary from the video description:
Sadly your comments reek of dishonesty. I am not sure who you are intending to convince with made up facts and goal posts.
Which is exactly fits in with what I have stated (for the 3rd time now):
But sure, you do you. Don’t forget to add “do not make mistakes” in your agents.md/skills.md/etc files and that have that RM790 Claude subscription handy. You’re going to need it.
adkilla,
I’d be very glad if you don’t skip some of the parts, cut some pieces, and just reach a predetermined conclusion.
I don’t disagree with most of what you said, however I don’t think you ever actually read what I tried to convey.
So, I’ll have to repeat, maybe quickly this helps.
1 – AI will not “magically” make bad programmers better ones., It won’t even help mediocre ones too much, if used incorrectly.
2 – AI is immensely useful to senior developers. It makes the output much better.
To me when organisations like Atlassian, once upon a time the angels of the IT sector, when they start AI Washing redundancies you know the writing is on the wall! How does someone who champions the environment, nature and people on one hand end up so morally barren that code replaces coders? Money!
I can only hope that when it all doesn’t work out after the AI bubble pops, most devs would have moved on to other fields and not look back. The tech industry needs a painful reminder of the dotcom bubble and pay for it hard.
adkilla,
I cite the dotcom bubble as a model for an AI bubble too. The thing is about the dotcom bubble though is that is didn’t represent the end of the market at all but rather massive consolidation and growth at the very top at the expense of everyone else. The same pattern could play out with AI.
Alfman,
There’ll always be winners/losers and eventual market consolidation. Many of the ‘new’ businesses and roles created during the dotcom hype eventually failed and only the tangible ones remained. At the very least, we could hope that when this happens, the market moves on and real innovation/opportunities can resume again.
adkilla,
Indeed. I just worry that by not having our own horse in the race promoting FOSS interests, it could end up producing winners that are much worse for FOSS and our independence. By the time enough people recognize this though, it’s likely going to be much more difficult for new FOSS friendly challengers to break into the market.
For employers, technology that automates “human” tasks is innovation. Even if all consumer AI applications fell through, the cash cow for AI was always going to be bringing automation to the business space. For industries like Hollywood, cutting edge AI is already producing impressive results. They’re just tipping their toes in today, but over time it seems almost inevitable that studios will outsource more work to VFX companies that use AI over more expensive VFX companies that don’t. It’s just the latest evolution, like CGI taking over for practical effects in the past. Do some people miss practical effects? Absolutely! Is it likely to come back though? No, money won.