Amazon Web Services on Tuesday announced three new AI agents it calls “frontier agents,” including one designed to learn how you like to work and then operate on its own for days.
Each of these agents handle different tasks such as writing code, security processes like code reviews, and automating DevOps tasks such as preventing incidents when pushing new code live. Preview versions of the agents are available now.
Perhaps the biggest and most interesting claim by AWS is its promise that the frontier agent called “Kiro autonomous agent” can work on its own for days at a time.
Kiro is a software coding agent based on AWS’s existing AI coding tool Kiro, which was announced in July. While that existing tool could be used for vibe coding (which is really just prototyping), it was intended to produce operational code, or software that would be pushed live. To make reliable code, the AI must follow a company’s software-coding specifications. Kiro does that through a concept called “spec-driven development.”
As Kiro codes, it has the human instruct, confirm, or correct its assumptions, thereby creating specifications. The Kiro autonomous agent watches how the team works in various tools by scanning existing code, among other training means. And then, AWS says, it can work independently.
“You simply assign a complex task from the backlog and it independently figures out how to get that work done,” AWS CEO Matt Garman promised when introducing the new product during his keynote at AWS re:Invent on Tuesday.
“It actually learns how you like to work, and it continues to deepen its understanding of your code and your products and the standards that your team follows over time,” he said.
Techcrunch event
San Francisco
|
October 13-15, 2026
Amazon says Kiro maintains “persistent context across sessions.” In other words, it doesn’t run out of memory and forget what it was supposed to do. It can therefore be handed tasks and work on its own for hours or days, Amazon promises, with minimal human intervention.
Garman described a task like updating a bit of critical code used by 15 bits of corporate software. Instead of assigning and verifying each update, Kiro can be assigned to fix all 15 in one prompt.
To complete the automation of coding tasks, the cloud provider developed AWS Security Agent, an agent that works independently to identify security problems as code is written, tests it after the fact, and then offers suggested fixes. The DevOps Agent rounds out the trio, automatically testing the new code for performance issues, or compatibility with other software, hardware, or cloud settings.
To be sure, Amazon’s agents aren’t the first to claim long work windows. For instance OpenAI said last month that GPT‑5.1-Codex-Max, its agentic coding model, is designed for long runs, too, up to 24 hours.
It’s also not totally clear that the biggest hurdle to agentic adoption is the context window (aka the ability to work continuously without stalling out). LLMs still have hallucination and accuracy issues that turn developers into “babysitters,” they say. So developers often want to assign short tasks and verify quickly before moving on.
Still, before agents can become like co-workers, context windows must grow bigger. Amazon’s tech is another big step in that direction.
