AWS Unveils ‘Frontier Agents’ That Can Code, Test and Diagnose on Their Own
Kiro, Security and DevOps agents carry long memory, hunt bugs and vulnerabilities, and trace outages across complex systems.
Topics
News
- Agentic AI Push Spurs Calls for Human Oversight, Clear Rules
- India Plugs Into Pax Silica as AI Supply Chains Realign
- India AI Impact Summit 2026: Did India’s Tech Push Just Take a Hit?
- When Autonomous AI Tests the Limits of Oversight
- UN Bets on Science to Steady AI Governance
- Legacy IT Leaders Say India’s AI Future Lies in Talent, IP
AWS, Amazon’s cloud-computing arm, has launched a set of new AI “frontier agents” designed to take on broader, more autonomous roles across the software development lifecycle, from coding and security to operations.
The announcement follows internal research at Amazon that found development teams were most effective when agents were given goal-driven tasks, allowed to run multiple workflows in parallel, and permitted to operate independently for longer stretches.
The new agents, dubbed Kiro autonomous agent, AWS Security Agent and AWS DevOps Agent, are designed to run for hours or days on their own, handling multi-step tasks that span coding, code review, security analysis and incident response.
Kiro acts as an autonomous software engineer. It keeps a persistent view of codebases, issues and pull requests, learns from feedback, and can triage bugs, suggest fixes and orchestrate multi-repository changes. Developers can assign work from GitHub or Jira, then review its output as proposed edits or pull requests rather than live code merges.
AWS Security Agent serves as a virtual application-security reviewer. It reads design documents, scans code for risks, flags insecure changes in pull requests and proposes remediation. It can also run on-demand penetration tests at the pace of a team’s deployments. Early testers such as SmugMug said the agent surfaced vulnerabilities that conventional tools did not pick up.
The DevOps Agent focuses on reliability and incident response. It analyzes telemetry from services such as CloudWatch, Datadog, and Splunk, maps dependencies across systems, and identifies root causes during outages.
It also provides recommendations based on past incidents. Commonwealth Bank of Australia reported that the agent diagnosed a complex network and identity issue in under 15 minutes during testing.
AWS has positioned the trio as a step beyond prompt-based copilots toward agents that stay engaged for long stretches, carry broad context and operate more like embedded team members across development, security and operations.
