▲ 198 points
back
97 comments
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.
> 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 :')
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.
- claude is relatively new, was only whitelisted about 1-2 months ago. kiro is still de-facto, a lot of ai tools use its acp as a back-end.
- there are no token budgets
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.
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
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.
> 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
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?
"Sign up for free access to this post".
There should be a rule about this kind of posts. If there isn't already.
> If there isn't already.
There is![0]
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.
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.
It’s the same as posting paywalls articles. There’s no need to make a new rule. If the users can’t read it, the conversation will simply devolve into title commenting.
Amazon employee here.
- Can confirm that channel exists and it's great. Rofl every day for all the stupidity happening in the company.
- Despite many seeing Amazon as outsider for AI, it's actually a great place to work (if not fired). We have free access to all frontier models, including today's Mythos release for all employees.
- Every model available on Bedrock is available for free with no caps. I think it's because these models are deployed on trainium chips which is essentially free for Amazon.
- Claude code and Codex are available unlimited for employees.
- For this one I will be bitten, but I actually like Kiro CLI. It's powered by the same Anthropic/OpenAI models. It's a bit easier to understand for me than CC and surprisingly does a better job at writing research papers.
- Amazon Quick, alternative to Claude Cowork, also has some nice features like in-browser presence so you don't have to copy/paste content from the web.
- I personally haven't seen a push to use AI at all costs. Yes, there are many internal tools being developed (e.g. meshclaw) but no one tracks their adoption for redundancy purposes. I think leaders are pretty consistent with what they communicate externally (e.g. hiring of juns).
Reminds me of Memegen from Google, the one thing I miss about working there.
Memes were top shelf; absolutely masterclass
I thought the channel is strictly for being sad that Kinesis in IAD is down again.
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.
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.
Then how do they get promoted?
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.
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.
> 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.
Sidebar: I love how Amazon basically gave up on AWS Chime for Slack. Even dogfooding their own messaging product didn't work.
IDK how the Chime team managed to make such a garbage product. Like, somehow they managed to make something even worse than Teams with a UI that made Webex look modern. I know there are good product people inside Amazon, so IDK what combination of incentives resulted in Chime.
Unless something has changed recently, Slack uses AWS Chime SDK for audio/video, so at least the core technology is great.
The Chime team is still busy. They released a new version just two weeks ago!
When AWS had meetings with us we’d insist on setting up the call ourselves to avoid using the steaming pile of garbage that was Chime. AWS folks confided that they didn’t mind and weren’t thrilled about being forced to use dogfooded stuff that literally seemingly no customers wanted to buy.
> 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.
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.
Perhaps Kiro has yet to master reading text in images.
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.)
On one hand, they don't want to piss off all their employees with big brother culture.
To some extent, it's less offensive to mock your own stuff the bigger you are. In my limited big corp experience, everything was so big and convoluted it was never anyone's fault (150 teams worked on <thing>) something sucked and there were always 25 more teams working to fix it anyway. In some ways, it was actually helpful because people would stay engaged trying to improve things "on a big scale"
Suppressing systematic issues makes it easy to carve out a niche pretending to look useful while doing nothing (which is ultimately a waste of money)--big corp already have plenty of those
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.
As an individual contributor I sell my time, not my soul.
Even before the internet, employees were making these kinds of criticisms at lunch, at the water cooler and during their car pools.
If the company forbids it on visible internal channels, it will just pop up on external, private channels. With less corporate control and less corporate visibility and more leaks.
Another one of these 404 Media bait posts, okay, we got it, there's internal staff memes.
Related:
Google employees internally share memes about how its AI sucks
agree. im already beginning to see this domain as indicator of trash quality post
why would hackers be interested in this
You cant deliver and have quality engineering teams, while you spent the last 15 years treating your employees under a frugal way, and being the biggest H1B IT employer in the US.
Hardly newsworthy, everything is mocked on the internal slack lol.
Are they don’t anything useful with AI? I would have figured it could help their operations.
Archive link? I'm not making an account to look at their memes, come on
Any text forums these days has the grokenspiels played by sloppenheimers. It feels like a literature cross between Punk'd and Crank Yankers but stars clanker wankers rather than TV and movie celebrities.
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