Amazon Web Services has introduced three new artificial intelligence agents it calls “frontier agents,” including an autonomous coding system named Kiro that the company says can operate independently for hours or even days, handling complex software tasks with minimal human input.
Three agents previewed for coding, security, and DevOps
AWS announced the new agents on Tuesday and said preview versions are available now. The three agents are designed for different technical functions: writing and maintaining code, handling security processes such as code reviews, and automating DevOps operations including testing and deployment checks.
The agents include the Kiro autonomous coding agent, the AWS Security Agent, and the DevOps Agent. Each is intended to work with existing software development workflows rather than operate as a standalone tool.
Kiro autonomous agent builds on earlier AWS coding tool
The Kiro autonomous agent is based on AWS’s earlier AI coding tool, also named Kiro, which the company announced in July. The original version was positioned to generate production-ready code rather than short prototypes. It relied on a system AWS describes as “spec-driven development,” in which software is created based on explicit coding standards and specifications.
As Kiro produces code, a human developer instructs, corrects, or validates its assumptions. These interactions become part of the specification the system follows. The new autonomous version expands this approach by observing how teams work across different tools, scanning existing code, and using that information as training input.
AWS chief executive Matt Garman said during his keynote at AWS re:Invent on Tuesday that users can assign Kiro a complex task directly from a development backlog and allow it to determine how to complete the work on its own. He said the agent continues to refine its understanding of a team’s codebase, products, and internal standards over time.
Persistent context and multi-day operation
Amazon said Kiro maintains what it calls “persistent context across sessions,” meaning it does not lose its running memory when sessions end. This allows the agent to continue working across extended periods instead of restarting with each interaction.
According to AWS, the agent can work independently for hours or days with limited oversight. Garman gave an example in which Kiro could update a critical piece of software code used across 15 separate internal systems. Rather than assigning and validating each update individually, the agent could be instructed to fix all 15 within a single task.
Security and DevOps agents handle testing and operations
The AWS Security Agent is designed to operate independently to identify security issues as code is written. It can test code after development and suggest fixes when it detects vulnerabilities.
The DevOps Agent focuses on deployment and operational checks. It automatically tests new code for performance issues and evaluates how it interacts with other software systems, hardware configurations, and cloud environments.
AWS positioned the three agents as complementary tools covering different stages of the software lifecycle.
Comparison with other long-running AI agents
AWS is not the first company to promote long-running AI agents for coding. OpenAI said last month that its GPT-5.1-Codex-Max model was designed for extended runs of up to 24 hours.
Industry discussions around agent-based systems continue to focus on challenges such as hallucinations and accuracy limits in large language models. Developers often assign narrowly scoped tasks and verify output quickly due to these limitations.
AWS presented its new agents as capable of sustained work across longer time windows, with Kiro designed to retain task context across multiple sessions.
Featured image credits: Wikimedia Commons
For more stories like it, click the +Follow button at the top of this page to follow us.