AWS puts AI agents to work on truly autonomous software development
by Mike Wheatley · SiliconANGLEAmazon Web Services Inc. is using its annual AWS re:Invent 2025 extravaganza this week in Las Vegas to show how it’s putting artificial intelligence agents to work in enterprise environments.
The cloud giant is moving on from the days of passive generative AI chatbots, building highly autonomous and massively scalable digital workers that can collaborate with humans and each other, and work under their own steam, potentially for days on end, with only minimal supervision.
At the core of this vision is Amazon’s full-stack architecture, including its powerful new Nova large language models, its custom Trainium3 AI processors and an agentic runtime that eliminates all of the hassles of getting agents up and running and ensuring they’re secure.
In an exclusive interview with SiliconANGLE, AWS Chief Executive Matt Garman said autonomous agents are the company’s top priority in terms of AI development. “The next 80% to 90% of enterprise AI value will come from agents,” he said.
Amazon’s agentic push will be led by a newer, more sophisticated class of AI agents that it calls “frontier agents,” which promise to deliver a “step-function change” in what digital workers can achieve. Rather than just assisting human workers like today’s AI coding assistants, for example, they can undertake complex projects independently, in the same way as people do. The company is kicking things off with three new frontier agents – Kiro, AWS Security Agent and AWS DevOps Agent, each one focused on a different aspect of the software development lifecycle.
Kiro Autonomous Agent
While AI coding tools have already performed miracles in terms of accelerating productivity, they have also created a lot of friction for human developers. These days, many coders find themselves having to act as the thread that holds multiple AI agents together, constantly rebuilding context when switching tasks, manually coordinating cross-repository changes and stitching together information from different tickets, pull requests and chats so they can properly guide them. In effect, it means developers have become coding agent overseers, but that prevents them from focusing on more creative priorities, which was the whole point of AI automation in the first place.
With Kiro, AWS is helping developers eliminate this fraction and ensure that agentic processes keep moving along nicely without constant human oversight. As a truly autonomous agent, Kiro is uniquely able to maintain persistent context across sessions and will continuously learn from, and remember, pull requests and human feedback, so it gets better over time.
It’s able to handle multiple tasks at once, ranging from triaging bugs to improving code coverage, across multiple code repositories. Users can ask it questions, describe the task they want it to do in natural language, or just tell it to work through the tasks in their GitHub backlog, and it will do so all by itself, figuring out what it needs to do without any human input, Amazon said. Changes will be shared as proposed edits and pull requests, ensuring the developer maintains overall control.
AWS Security Agent
While Kiro handles the grunt work, the AWS Security Agent is all about oversight, proactively identifying risks and taking steps to mitigate them once an issue has been identified. Whereas existing security agents only provide generic recommendations, AWS Security Agent offers tailored guidance throughout the software development lifecycle and can perform comprehensive testing at any stage.
The agent works by proactively reviewing design documents, scanning pull requests and comparing these with organizational security rules and its list of common vulnerabilities. All the user has to do is define their security standards once, and the agent will automatically validate them across every application that’s hosted on AWS.
In addition, it can also perform penetration testing on demand so security issues don’t hold back development velocity, Amazon said. It validates any problems it finds and then generates remediations to fix those issues, before applying them once developers give it the go ahead to do so.
AWS DevOps Agent
The final piece of the puzzle, AWS DevOps Agent works to maintain the underlying infrastructure that distributed applications depend on – monitoring microservices, cloud dependencies and telemetry to gain a comprehensive understanding of its behavior.
The company said AWS DevOps Agent will be on call 24/7, ensuring it’s ready to respond instantly the moment any incidents occur. It draws on its knowledge of the customer’s applications and its relationship with the various infrastructure components to identify the root cause of any system failure or performance issues quickly, employing observability tools such as Amazon CloudWatch, Dynatrace, Datadog, New Relic and Splunk. By mapping each application’s resources with its telemetry, code, and deployment data, it can rapidly pinpoint root causes and accelerate resolution times, while delivering fewer false alerts, Amazon said.
With AWS DevOps Agent, teams can also move away from reactive firefighting and became much more proactive, using it to analyze the cause of historical incidents and prevent them from recurring. By learning from these experiences, it can offer targeted recommendations to enhance observability, infrastructure optimization, deployment pipeline enhancement and application resilience, the company promised.
Bringing autonomous software to life
Garman said all three frontier agents can be set up once and then be left to work for weeks on end, scaling up or out as required. As they do this, they will learn each customer’s preferences over time, improving their performance the more they’re used. “Three to six months in,” Garman said, “these agents behave like part of your team. They know your naming conventions, your repos, your patterns.”
Over time, Amazon expects these frontier agents to help organizations shift to an “agentic culture” where AI is no longer just an assistant, but an extension of their human teams. Ultimately, it sees its frontier agents delivering outcomes autonomously throughout every step of the development lifecycle, transforming how software is built, secured and operated.