Marc Brooker, an AWS VP and distinguished engineer, leads Velocity Labs, a small team inside Amazon’s cloud unit. (Amazon Photo)
Last month, a small team inside Amazon Web Services called Velocity Labs went from concept to the public release of a new open-source AI model in less than a week.
The group, whose existence hasn’t been previously reported, was formed in April. Velocity Labs has about 35 people, most of them in Seattle. The idea is to think beyond Amazon’s normal product-oriented approach, building quick prototypes in areas of AI that the company sees as underexplored and increasingly important to customers.
“We have noticed how the open-source community, and some of our competitors, experiment with new ideas, and deliver initial products fast,” the company says in job listings . “With Velocity Labs we are building an organization that can move faster, explore more, and make sure AWS is at the cutting edge of innovation in agentic AI.”
Inside the group, scientists and engineers work together, and they’re encouraged to cross over: engineers engage in research, and scientists do engineering work.
Velocity Labs is an example of the ways Amazon is trying “to create organizations that we think are going to be better suited to working with this new set of tools,” said Marc Brooker, an AWS vice president and distinguished engineer who leads the team.
For the vast majority of Amazon’s developers, he said, AI agents now write almost all the code they would previously have written by hand.
Redefining the role: The team is part of Amazon’s larger rethinking of software development for the AI era, laid out in revised guidelines that the company’s engineering leaders sent to its developers earlier this year.
A letter that accompanied the guidelines, published last week under the byline of Terk Terkowitz, a VP and distinguished engineer in Amazon’s stores business, advises engineers to learn AI now and to stay close to their customers and products.
Amazon has never published its underlying role guidelines for software engineers, and it declined to provide a copy of the revised version for this article. Brooker, one of the authors of the accompanying letter, said in an interview that the update emphasizes measuring engineers and teams by outcomes rather than inputs.
“We had seen these anti-patterns in the industry of folks really focusing on how many tokens are you spending, how many lines of code are you committing,” Brooker said, explaining that those kinds of measures don’t equate to customer value.
Asked whether AI use is part of performance reviews or promotions, Brooker said, “Not directly in a metrics sense.” It could factor in, he said, to the extent that it makes someone a stronger performer.
The letter doesn’t dismiss usage metrics entirely. Amazon has found that they “correlate well with the outputs that matter” and expects that to continue “for some time,” it says.
But what’s important, the letter says, is to use them “with care and judgement.”
Amazon engineers have described pressure to use AI tools. Three developers told The New York Times last year that managers were pushing them to use AI, raising output goals and becoming less forgiving about deadlines, with some coders saying their jobs had begun to resemble warehouse work. Amazon has set a goal for more than 80% of its developers to use AI each week, the Financial Times reported .
In May, Amazon shut down an internal leaderboard called KiroRank — which ranked employees by how much they used Kiro , the company’s AI coding tool — after finding that some were using the tool excessively just to climb the rankings, the FT reported.
Asked whether that episode was part of the impetus for the revised guidelines, Brooker said he didn’t recall the leaderboard or the details surrounding it.
@import url(‘https://fonts.googleapis.com/css2?family=Roboto+Slab:wght@500&display=swap’);@media (max-width:768px){.gw-pq{display:none!important}}
“If your work does not benefit from current AI tooling, that is a finding, not a failing. Document why, share it with your community of peers, bring it to your manager, and work to find solutions.”
Amazon engineering leaders
Letter accompanying the revised SDE role guidelines
How the work has changed: Two years ago, most Amazon developers were still writing their code by hand, Brooker said. Today, “for the vast majority of our developer population, almost all of their implementation code is agent generated rather than human written,” he said.
“That is a massive change in the way of working for software engineers,” he said.
Brooker said he used to spend much of his time in a traditional code editor or the command line, and now spends a lot more of it working with agent tools such as Kiro, Kiro Crew and Anthropic’s Claude Code, calling it “a complete change of tools.”
For the company, the transition hasn’t been seamless. In March, Amazon acknowledged that an engineer who followed inaccurate advice from an AI agent contributed to one of several outages on its retail site, though the company disputed a Financial Times report that junior and mid-level engineers would need senior engineers to sign off on AI-assisted changes.
Amazon also pushed back on an earlier FT report that a 13-hour disruption to an AWS service in December happened after engineers let its Kiro AI coding tool make changes, and the tool decided to “delete and recreate the environment.” Amazon blamed user error — a misconfigured role — and said it has since added safeguards, including mandatory peer review for production access.
Still, the letter acknowledges that today’s AI tools won’t help in every situation, and it says that’s OK as long as engineers can explain why.
“If your work does not benefit from current AI tooling, that is a finding, not a failing,” the letter says. “An engineer who can explain where AI does not yet help their work is demonstrating the judgement we are asking for.”
Inside Velocity Labs: The group is part of the agentic AI division led by Swami Sivasubramanian, which has organized much of its work around small teams of six to eight people, as GeekWire reported in June . Velocity Labs also includes the developers behind Strands , AWS’s open-source toolkit for building AI agents.
The model that Velocity Labs released last month, Strands Decider 2B , is designed to make quick, narrow decisions inside AI agents, such as which tool to use or where to route a request. It’s based on Qwen, an open model from Alibaba, and was released Oct. 1 along with the recipe used to train it.
Brooker built the original version on his own GPU at home after the startup TypeSafe AI unveiled Jev, a similar decision model, on Sept. 15, he wrote on his blog . “Every line of code was written by an agent,” he wrote, though he tried to make sure “the core ideas and insights were mine, or at least I understood them.”
Brooker described the team’s turnaround as less than a week from concept to public release. OpenAI announced a similar hosted service, its Decisions API, the same week, TechCrunch reported .
Brooker credited the team’s speed to its small size and a high level of trust: each person takes ownership of a piece of a project and asks for help when needed. He expects Amazon to create more teams like it, though not across the whole company.
Impact on jobs: Amazon has cut roughly 30,000 corporate jobs since late 2025. CEO Andy Jassy has said he expects AI to shrink the company’s corporate workforce over time.
At the same time, AWS CEO Matt Garman said in April that Amazon is “hiring just as many software developers as we ever had,” Business Insider reported . An Amazon spokesperson told the publication that more than 11,000 interns and early-career software engineers would join the company globally this year.
Asked whether AI will mean fewer engineers at Amazon, Brooker said, “I don’t think it changes the number of engineers.” The company wants engineers who take ownership, are curious and can learn quickly outside their areas of expertise, he said.
AI isn’t the right answer for every problem, Brooker said.
If the right tool is “an AI agent powered by a cutting-edge model, fantastic,” he said. However, he added, if it’s “three lines of Python code, that’s great too.”
