Amazon Tells Its Engineers: Use Our AI Coding Tool 'Kiro'

"Amazon suggested its engineers eschew AI code generation tools from third-party companies in favor of its own ," reports Reuters, "a move to bolster its proprietary Kiro service, which it released in July, according to an internal memo viewed by Reuters." In the memo, posted to Amazon's internal news site, the company said, "While we continue to support existing tools in use today, we do not plan to support additional third party, AI development tools.

"As part of our builder community, you all play a critical role shaping these products and we use your feedback to aggressively improve them," according to the memo.

The guidance would seem to preclude Amazon employees from using other popular software coding tools like OpenAI's Codex, Anthropic's Claude Code, and those from startup Cursor. That is despite Amazon having invested about $8 billion into Anthropic and reaching a seven-year $38 billion deal with OpenAI to sell it cloud-computing services..."To make these experiences truly exceptional, we need your help," according to the memo, which was signed by Peter DeSantis, senior vice president of AWS utility computing, and Dave Treadwell, senior vice president of eCommerce Foundation. "We're making Kiro our recommended AI-native development tool for Amazon...."

In October, Amazon revised its internal guidance for OpenAI's Codex to "Do Not Use" following a roughly six month assessment, according to a memo reviewed by Reuters. And Claude Code was briefly designated as "Do Not Use," before that was reversed following a reporter inquiry at the time.
The article adds that Amazon "has been fighting a reputation that it is trailing competitors in development of AI tools as rivals like OpenAI and Google speed ahead..."

Seems reasonable

[houstonbofh]

Makes sense to me. Why give their code and feedback to competitors? You want to keep your internal stuff internal.

Re:

[fuzzyfuzzyfungus]

It seems reasonable; but also like something that should really spook the customers.

It seems to be generally accepted that junior devs start out as more of an investment than a genuine aid to productivity; so you try to pick the ones that seem sharp and with it, put some time into them, and treat them OK enough that they at least stick around long enough to become valuable and do some work for you.

If that dynamic is now being played out with someone else's bots, you are now making that investment in s

Re:

[allo]

I think many programmers got good through "learning by doing", which involves for example programs that are mostly StackOverflow snippets pasted together to create an app. Starting from that they get better until they finally trash (or archive) their first app and rewrite it themselves. Now they will combine LLM outputs, but they are still exposed to working code and get to learn from it.

Re:

[ArmoredDragon]

We do exactly this where I work (I'm part of the team who enforces that) and for exactly this reason. However in our case, until a few months ago, use of these tools was nearly completely banned with case-by-case exceptions for individual users, unlike Amazon. We only recently made one exception for just one AI tool where we have a particular arrangement that guarantees that nothing within our instance of it can be "learned" and regurgitated elsewhere, and anybody may use it without needing any policy excep

Re:

[allo]

Eat your own dogfood. I would be more worried when they create tools and don't use them themselves.

BS Krebs Cycle

[Tablizer]

1. Amazon orders own devs to use their AI tool Foo
2. Devs reluctantly follow orders
3. Foo usage goes up
4. Amazon brags about high Foo usage in ad
5. Suckers see ad, buy Foo
6. Profit!

Re:

[Iamthecheese]

There's an element you're not considering. These engineers are eating their own dogfood, poor quality will lead to product changes, and so expected quality must rise.

Re:

[Tablizer]

Most devs at Amazon are probably not working on AI engines.

Re:

[93 Escort Wagon]

"Hey Kiro, I'm finding that you really suck at generating code. Please rewrite your internal codebase so you suck less."

Re:

[allo]

Devs having to use the AI engine will write tickets for devs building the AI engine. If all of that happens inside Amazon it is way more efficient than having to communicate with a contractor. When buying from another company, you have different incentives. The workers at Amazon still want the best AI, while the other company wants to deny as many requests as possible to keep things cheap and keep their own development direction instead of having Amazon devs influence it. In-House systems have a clear advan

Re:

[Afell001]

There is a benefit to using other AI tools too - because exclusive use of internal tools can cause ideological inbreeding. It would be better to encourage broad use when prototyping and try to bring the best of the experience home to the tools in house.

There does not seem to be a bright future for them

[gweihir]

All these coding "assist" tools suck regarding functionality, and they suck worse regarding security. Expect Amazon to get flaky and then get hacked in the not to distant future.

In other news, throwing good money after bad or going "all in" on bad ideas are not a successful business strategy.

Re:

[Anonymous Coward]

I mean, you're right. But you forgot the only thing that matters: stock price. Amazon is using AI? Pump that stock! They're using an internally developed AI? Think of the more hyuger cost savings! Double pump! Just gave 38 billion to another company with a ineffective product. Chump change. Now we have to spend another 38 billion on our own ineffective product? Gotta spend money to make money. As long as the number goes up. Amazon, like Bender, can promise you whatever you want.

Re: There does not seem to be a bright future for

[liqu1d]

That's my experience too. 90% of the code I've got out of it takes longer to debug than if I wrote it myself however way too many people claim to experience the opposite. How can our experiences differ so vastly.

Re:

[gweihir]

Indeed. I think the experience is not different, the perception is. First, a lot of the people that really like AI-produced code probably never really debug it carefully. Run it on some test data. Works? Use it! For code security, this will be worse.

Also, people think AI coding assistants make them faster, when in reality it usually makes them slower: https://mikelovesrobots.substa... [substack.com]

Bottom line is, as so often, people in awe of new tech without understanding its limitations. And then bashing everybody that

Re: There does not seem to be a bright future for

[LindleyF]

You're using it wrong. If you are trying to vibe code at such a high level that ideas like "security" matter, you're going to have a bad day. The ideal case is when you have an existing codebase that requires a tweak. You already know just about what you want to do to it. Doing it yourself would take an hour or two. Using an AI might take a minute or two. But since you know exactly what you wanted, it's easy to verify the result. Plus, many of them are pretty good at "now add tests for that".

Re:

[Afell001]

Where I've found it useful is on the drudgery that I usually end up clobbering myself with over the last leg of a sprint - like unit tests or rough documentation. I can trust that the AI is going to do a pretty decent job developing the unit tests - hard to screw those up, really, and it doesn't take long to review them for accuracy. I treat it like the junior dev you don't trust to code their way out of a wet paper bag. The documentation part is very useful, because if the AI can't make heads or tails of w

Re: Did they tell anyone how to pronounce it?

[sethadam1]

It's Kee-row

Priority

[backslashdot]

Is the priority good code or emotional "don't use the competing product"? Remember, the Apple Macintosh SE was designed using a Cray supercomputer. The Cray supercomputer had been designed on an Apple Macintosh.

If you make slow bicycles, and need to win a bicycle race .. would you ride your own product knowing it may cause you to lose? If your dog doesn't like your dog food, then don't feed it to him.

(This is obviously not applicable and different than a team that is testing its own product.)

It's ready

[sethadam1]

I've tried all of these AI code editors: Zed, Windsurf, Cursor, and of course VSCode and ultimately landed on Kiro. I've been using it for a few months as a paid user, as I lucked into early access and then coughed up payment when it went live. It's by far the best in my experience in getting it right most of the time. I struggled with even getting Windsurf to reply reliably - interesting since they all really use the same services on the backend. Kiro also looks great on the Mac, where some of the others really feel like poorly integrated Electron apps. So far I'm enjoying it and I'm glad Amazon engineers will have to use it, because that will probably lead to further improvements.

AGI nowhere to be seen

[evanh]

Just another search engine come database that requires lots of examples to copy.

Formerly known as Amazon Q

[Mean Variance]

My work has been open to using AI for coding and deeper use cases. We have CoPilot and a limited Gemini. Q (now Kiro) was quietly introduced a few months ago. If you are a Saas on AWS, it's really helpful for contextual cases involving the cloud service stack because it can utilize whatever permissions are available. Example: evaluate my cold starts for lambda fancy-service-abc in region us-west-2. It's much faster than slogging through the online console or if you remember all the CLI for aws -- cheers! -- I don't because it's not an everyday thing for me.

Now I'm using it for code scenarios but needs good agentic design. That's the thing, all this AI isn't worthless and it's not yet going to replace experienced developers. I can scaffold code so much faster with my self designed agent which takes time to sort out but has been paying off.

AI - any of it - is just a another tool to build software, and create things.

Calling dog owners ...

[RockDoctor]

(With or without a cat-staffing job too.)

What is the normal result of forcing a dog to eat it's own vomit?