Amazon Web Services is moving deeper into agentic software development with three new AI agents it calls "frontier agents." The set includes Kiro autonomous agent for coding work, AWS Security Agent for security review and suggested fixes, and DevOps Agent for performance and compatibility checks.
The preview versions are available now. AWS is positioning the agents around a bigger idea: AI systems that do not just answer a prompt, but stay with a software task long enough to understand how a team works and carry that task forward.
What AWS Announced
The three agents are aimed at different parts of the software delivery process. Kiro autonomous agent is the coding system. AWS Security Agent is designed to find security issues while code is being written, test after the fact, and recommend fixes. DevOps Agent focuses on testing new code for performance issues and compatibility with other software, hardware, or cloud settings.
AWS describes these as preview products, not finished replacements for engineering teams. The company’s strongest promise is attached to Kiro: it says the agent can operate independently for hours or days with minimal human intervention.
That matters because most AI coding tools still work best when a developer keeps the task narrow. The developer asks for a change, reviews the output, corrects mistakes, and then moves to the next step. AWS is trying to stretch that pattern into a longer workflow.
Why Kiro Is The Main Focus
Kiro autonomous agent builds on AWS’s existing AI coding tool Kiro, which was announced in July. The earlier Kiro could be used for vibe coding, which the source describes as prototyping, but it was also intended to create operational code that could be pushed live.
For that kind of work, the agent needs more than a prompt. It needs to follow the company’s coding specifications and fit into the way a software team already builds, reviews and ships code. AWS says Kiro approaches that through "spec-driven development."
In that workflow, the human user instructs, confirms or corrects the AI’s assumptions while Kiro writes code. Those interactions help create specifications. The autonomous version then watches how the team works across tools, including by scanning existing code, and AWS says it can use that understanding to work independently.
AWS CEO Matt Garman introduced the product during his keynote at AWS re:Invent on Tuesday. He described a backlog-style workflow in which a team assigns Kiro a complex task and the agent figures out how to complete it.
Persistent Context Is The Big Claim
The phrase that explains AWS’s pitch is "persistent context across sessions." In plain language, Amazon says Kiro does not lose track of the task or forget the relevant background when work stretches beyond one interaction.
That is the feature AWS connects to longer-running autonomy. If an agent can keep the relevant context, it can receive a larger assignment and continue moving through the work without constant restarts.
Garman gave an example involving critical code used by 15 bits of corporate software. Instead of asking a developer to assign and verify each update one by one, Kiro could be asked to fix all 15 in one prompt.
The practical promise is not just speed. It is reducing the coordination cost around repetitive or multi-part engineering work. If the agent can understand the codebase, the product, and the team’s standards over time, it can take on tasks that would otherwise require many small instructions.
The Security And DevOps Pieces
Kiro is the attention-grabbing agent, but AWS’s trio also covers two areas that determine whether code can safely reach production: security and operations.
- AWS Security Agent works independently to identify security problems as code is written, test code after the fact, and suggest fixes.
- DevOps Agent automatically tests new code for performance issues and checks compatibility with software, hardware, or cloud settings.
Together, the three agents reflect a fuller software lifecycle. One writes code, one checks for security problems, and one looks for operational issues before new code goes live.
That structure also shows why AWS is not only talking about code generation. The harder goal is connecting code-writing to the surrounding work that makes code usable in a production environment.
What Still Holds Agents Back
AWS is not alone in claiming longer work windows for AI coding tools. The source notes that OpenAI said last month that GPT‑5.1-Codex-Max, its agentic coding model, is designed for long runs too, up to 24 hours.
Longer context alone does not solve every adoption problem. The source points out that large language models still have hallucination and accuracy issues, which can leave developers acting as "babysitters." That is why many developers still prefer short assignments that can be checked quickly.
So the important question is not only whether Kiro can keep working for days. It is whether teams can trust the work enough to let the agent run that long without turning review into a larger burden later.
Still, the direction is clear. If AI agents are going to become more like software co-workers, they need to remember more, understand team standards better, and sustain work across longer sessions. AWS’s frontier agents are another step toward that version of AI-assisted development.