Amazon Web Services (AWS) revealed three new “frontier agents” on Tuesday. These AI systems aim to handle core engineering tasks with minimal oversight.
The most notable agent, Kiro, can operate on its own for days, according to the company. AWS released preview versions of all three agents immediately.
Each one targets a specific part of the software development process, such as coding, security reviews, and DevOps automation.
Kiro
Kiro builds on the original Kiro coding tool that AWS introduced in July. That earlier version supported rapid prototyping and also produced production-ready code.
It followed company coding rules through a method called spec-driven development. In this process, developers guide the AI by confirming or correcting assumptions.
Those confirmations then form clear specifications the system can follow.
The new autonomous version expands these capabilities by studying existing codebases and also observes how teams work across tools.
As it learns, it develops an understanding of a team’s coding style, product structure, and engineering standards.
During his keynote at AWS re:Invent, AWS CEO Matt Garman described how the autonomous agent handles complex assignments.
He said developers can give Kiro a high-level task from the backlog. The agent then figures out the steps required to complete that work.
Persistent Context
AWS emphasised Kiro’s “persistent context.” This means the agent does not lose track of instructions or details while it works.
Many AI tools forget earlier information during long sessions, which forces users to repeat guidance. Kiro, however, retains memory across long cycles.
Because of that, AWS claims the agent can work for hours or even days with little human involvement.
Garman offered an example. If a company needs to update critical code used in 15 separate services, Kiro can handle all updates through one prompt.
Typically, a developer would complete and verify each update manually.
Also read: OpenAI Brings Models to AWS
Security And DevOps Agents
1. AWS Security Agent
This agent independently checks code for security issues. It monitors code as it is written, tests completed code, and proposes fixes.
Therefore, it reduces the burden on security teams and helps prevent vulnerabilities from slipping into production.
2. DevOps Agent
The DevOps Agent tests code before deployment. It checks for performance problems and compatibility issues.
It also evaluates how new code interacts with other software, hardware, and cloud settings. As a result, development teams can resolve problems earlier in the release process.
Industry Momentum
AWS is entering a competitive space. Last month, OpenAI announced GPT-5.1-Codex-Max, an agentic coding model designed for long-running tasks of up to 24 hours.
Several companies now aim to extend context windows and reduce the need for constant developer supervision.
However, developers still report hallucinations and inaccuracies from large language models. These issues often force teams to assign short tasks and verify each step.
They also increase the need for oversight. Even so, longer context windows remain essential for agents that hope to function like co-workers rather than simple assistants.
