Top
Best
New

Posted by doener 2 hours ago

'Sloppenheimer:' Amazon Employees Mock the Company's AI on Slack(www.404media.co)
143 points | 68 comments
gavinray 2 hours ago|
I recently wound up in conversation with someone working at Amazon. Inevitably, the convo steered into AI/LLM's at some point.

Here is what I learned:

- AWS had an in-house LLM tool that was terrible they tried to use for a while

- A lot of them still use Kiro

- Claude Code is currently the de-facto standard

- They're in the process of getting some custom Codex variant that doesn't phone-home and is audited approved

- There's no mandated organizational standard for what exact tools to some, various teams have different levels of adoption and stacks

- No org-wide/team-wide conventions for Claude Code

- They do have token budgets

- There's an intenral push for something called "Agent Spaces" which was described to me as a sort of Lovable/Bolt-type thing if I understood it right

I can't validate all of this and I might have misremembered, but just in case anyone else finds it interesting.

pinkmuffinere 35 minutes ago||
> There's no mandated organizational standard for what exact tools to some, various teams have different levels of adoption and stacks

> No org-wide/team-wide conventions for Claude Code

Just for context, this pattern (different teams using different tools in different ways) is extremely normal within Amazon, and is intentional. These shouldn't necessarily be seen as a failure. Amazon likes to have multiple competing options they use for everything, and they constantly evaluate which option is best performing, like an A/B test. After a couple years they will pare away whatever performs worst, replacing it with a new option. This strategy definitely has it's disadvantages, but it is an intentional chosen pattern throughout the company.

Source: I worked there for 5 years, and painfully/tearfully remember the transitions chime -> slack -> teams and workdocs -> quip -> confluence :')

noisy_boy 1 hour ago|||
I am more curious about how much token budget they have. Here I have to beg my boss for more as if he is paying from his pocket or I am using it for my hobby projects (I am not). I guess time to go back to copy/pasting to chat and doing things by hand like a caveman.
garciasn 1 hour ago||
I don't understand this token budget shit. Why would anyone, in their right mind, not be using Claude Max? All of the engineers at our org are using 6.25x and several heavy non-developers are as well. The rest of the company with licenses are using 1.25x.

I have hit my 6.25x limit exactly once in the last quarter.

---

I realize that we will all eventually be forced to pay more for this and I have raised it as a real possibility to the org for budgeting scenario planning; however, for now, why would you pay by token when it's subsidized?!

e: I now understand; you can stop downvoting.

mrgaro 1 hour ago|||
In order to get enterprise agreement you need to pay per token for Claude.
garciasn 1 hour ago||
Ah; I didn't realize there was a 150 person cap to Team and I suppose paying out the ear is worth it for compliance audit. Makes sense in that regard.
disgruntledphd2 1 hour ago||||
> I realize that we will all eventually be forced to pay more for this and I have raised it as a real possibility to the org for budgeting scenario planning; however, for now, why would you pay by token when it's subsidized?!

Anthropic (and maybe OpenAI?) have gated all the important enterprise features behind API plus pricing in the last quarter or two.

dastbe 1 hour ago||||
because they’re not; anthropic is pushing enterprises to switch to API/token pricing
mrguyorama 1 hour ago|||
For the code I generate and the limited way I am using it, Claude Sonnet is reliable and good.

I hope that I can someday run something very much like it locally.

The moment that happens, the AI industry is essentially useless to me. I don't need some ultra expensive "Totally better" model that does the exact same thing.

sparrc 32 minutes ago|||
I work at AWS, we got access to Claude Code very recently, most people are using Kiro and that seems to be the defacto standard.

I have not heard of any nor run into any token budgets.

There is generally a lot of team-wide usage of CLAUDE.md/AGENTS.md, team-specific skills, oncall skills, etc.

Not as much org-wide, although my org has an MCP server built for helping with oncall.

Ancalagon 27 minutes ago|||
> There's an intenral push for something called "Agent Spaces" which was described to me as a sort of Lovable/Bolt-type thing if I understood it right

Everyone and their mother making vibe-app platforms for their company API's now

tyingq 1 hour ago|||
Would be really curious what the internal market share for Kiro is. Not a really good look for it if it's just smattering use here and there.
8note 35 minutes ago|||
assuming it hasnt changed, agent spaces is just a developer desktop running on actual aws rather than on internal hosts, and they figured out how to put vscode on it in a security approved way, including your employee credentials.

one of the things that allows is for adding mcps, skills, and various harnesses that are preconfigured to work out of the box.

i doubt its gotten out of the employee needing to sign in every couple hours

add-sub-mul-div 1 hour ago||
From the outside it always seems laughably circuitious compared to just learning skills ourselves.

Even if it did let me fill out TPS reports 20% faster, who even cares compared to all of this chaos?

dfee 1 hour ago|||
the problem is scale. there's a tension between an individual developing technical skills (transfer cost is high, slow, expensive) and developing agent skills (transfer cost is low, instant, free).

so, just like a manager manages employees, or you consult a contractor, agents are a way of getting leverage over a system.

that said, if you want to learn to play saxophone, you're free to do so. just note your personal endeavors may begin to look more like hobbies than marketable skills.

rustystump 4 minutes ago||
When the cost for this leverage is more than an employee the math stops mathing.

Additionally, for tech work. There is a tension about doing work and not knowing that output is correct or not. I have seen ai spit out thousands of lines of opencv code for a simple color lut. The person doing this had no idea what was going on. If they continued, the token cost and time waiting for agents spinning only goes up.

Yes, agents get smarter and cheaper but the above example replays over and over again even on crud apps. You still need to dev the skills and transfer costs for it to be effective.

wahnfrieden 1 hour ago|||
Shareholders want the company to figure out how to pay fewer employees, and pay them less by down-skilling them
stanac 2 hours ago||
"Sign up for free access to this post".

There should be a rule about this kind of posts. If there isn't already.

https://archive.ph/1YRCE

sarreph 1 hour ago||
> If there isn't already.

There is![0]

[0] - https://news.ycombinator.com/item?id=10178989

mirashii 1 hour ago||
Importantly, in that rule:

> Complaints about paywalls are off topic, so please don't post them.

cheshire_cat 1 hour ago|||
By that logic links to the FT and the WaPo also couldn't be posted. Which I guess is a fair position to take, but seems too limiting to me.
dminik 1 hour ago|||
What meaningful conversation can you have about an article you can read beyond the headline?
dylan604 1 hour ago|||
Just because you don't have access doesn't mean that people with access should be disallowed from discussing it. It just means you can't really discuss the article, but that's never stopped anyone from commenting before
rustystump 1 hour ago|||
Sadly, knowing most people, headline is rarely read past even when it isnt paywalled
functionmouse 1 hour ago|||
> seems too limiting to me.

articles restricting most users from reading them seems too limiting

ban all sites with paywalls/login walls including Twitter, NYT, FT, Business Insider, literally all of them

nosioptar 1 hour ago|||
For anyone who doesnt know: changing the x.com or twitter.com in a twitter link to xcancel.com usually works to view tweets without an account.

(If I remember right, some video links dont always work with xcancel.)

airstrike 1 hour ago||||
WSJ and Bloomberg offer "gift links" that paid users can share. The latter is only good for 7 days IIRC, however.
imglorp 1 hour ago||||
Seems pretty easy to have a bot automatically try to create an archive link and comment it.
incanus77 1 hour ago||||
Ok, anyone care to gather at a coffee shop so that we can discuss the print version?
ssl-3 1 hour ago||
You guys haven't been going to the meetings?
apparent 4 minutes ago||
No, my horse broke down so I've been stuck at the farm.
Forgeties79 1 hour ago|||
There are plenty of posts every day that don’t require you to get through a paywall - I personally just ignore them. I imagine there are only a handful of paywalled articles a week, and most of them end up having an archive link in the comments anyway
cyanydeez 1 hour ago||
Someone should point their LLM at creating a web plugin that just overlays a comment section & wiki style editing for archive and other similar websites. Then we can do the same thing LLMs are doing to invalidate primary sources.
cmiles8 1 hour ago||
Amazon’s attempts at AI tooling are just far too behind to be taken seriously. Kiro, Quick whatever, the Alexa updates. All just a hot mess. Amazon’s own employees appear to have abandoned Kiro en mass when allowed to just use Claude.

Amazon should just focus on being a utility compute provider. Anything they try to do on top of that is just consistently second rate.

computomatic 1 hour ago|
It’s all relative. Kiro is second to Claude Code, but Amazon isn’t really competing with Anthropic. They need something better than Microsoft/Github Copilot and that is a low, low bar.
giancarlostoro 1 hour ago||
I'm surprised with Amazon and Meta in terms of AI. Less surprised with Google, I think Google has a very specific niche they picked. If you really think about it, you don't have to be the absolute best, you just have to keep refining your model for efficiency and cost, and I think that's Google's true goals and secret sauce. Google will snowball into place. They're also used more than most people realize.

https://cloud.google.com/customers/qualia

Then there's Mark Z who is throwing away piles of money, but nothing to show for it other than letting people easily hack his social media sites? I really hope they never let an AI just send someone a link like that again? If you're at the point where a person is taking over an account, have a human review it, check for red flags like a VPN.

ako 1 hour ago||
I think Amazon is doing ok as the cloud where most customers run their LLM. I think a lot of companies are using e.g., Anthropic models on Bedrock so it lives inside their AWS cloud.
disgruntledphd2 1 hour ago||
> Then there's Mark Z who is throwing away piles of money, but nothing to show for it other than letting people easily hack his social media sites?

I mean, FB/Meta have been using lots of GPUs and compute for well over a decade at this point, they definitely are one of the few companies who can make use of relatively absurd amounts of these to drive revenue (i.e. improve ranking for both personal/organic and paid posts).

Whether or not they'll get a return from this wave is much more up in the air.

giancarlostoro 21 minutes ago||
What kills me more about them is they could easily build a coding model, purely focused on coding, and it could probably be competitive, instead they're wasting away at anime avatars or whatever Mark Zuckerberg needs for his metaverse utopia.
burger_moon 25 minutes ago||
The non-programmers at amazon I know are all just getting onto kiro. I guess there’s two versions of kiro(?) one aimed at devs and the other more web/mcp friendly. Ic and managers are still just wading into the waters and still haven’t learned that token dashboards are snake eating its own tail endgame for these groups.

From the outside it looks like a hot mess of many competing teams trying to become the chosen one for all of Amazon’s ai use. As a result it’s a confusing mess of tools that don’t last very long and creates tons of churn and confused employees who learn new tools as they're killed.

Definitely enjoying watching the chaos unfold from a distance at these trillion dollar businesses

bilsbie 24 minutes ago||
Are they don’t anything useful with AI? I would have figured it could help their operations.
nozzlegear 1 hour ago||
I've never worked in a big corporate environment, so I'm always surprised that these companies allow employees to mock their own products/managers/bosses.

(Not saying it shouldn't be allowed, just that it's surprising based on how controlling I'd expect a big corp like Amazon to be.)

TheCoelacanth 46 minutes ago||
It's a fool's errand to try to prevent criticism to that extent. It's mostly harmless and functions as a release valve for people's frustrations and helps to stop them from doing something more extreme than just complaining.

It's also possibly illegal to stop them. Employees in the US have a legal right to talk to each other about their working conditions and employers are not allowed to stop them. Most companies aren't above violating that law, but they want to save that for actual union-busting, not to stop people from sharing memes.

bobim 1 hour ago|||
As an individual contributor I sell my time, not my soul.
yesitcan 1 hour ago||
> I've recreated all the memes rather than share screenshots from the Slack channel in order to protect sources.

300 IQ move. Nobody will be able to trace it back to sources now.

05 24 minutes ago||
images/screenshots could have watermarks, recreations don't. Sources = not the people who created memes, sources here just mean people who sent the screenshots to the journalist.
Apocryphon 34 minutes ago||
Perhaps Kiro has yet to master reading text in images.
Perz1val 2 hours ago|
Archive link? I'm not making an account to look at their memes, come on
lucrbvi 1 hour ago|
http://archive.today/1YRCE
More comments...