AWS launches Kiro Crew, an autonomous agentic orchestrator for 24/7 code development

UPDATED 13:00 EDT / AUGUST 04 2026 AI AWS launches Kiro Crew, an autonomous agentic orchestrator for 24/7 code development by Kyt Dotson Amazon Web Services Inc. today launched Kiro Crew, an autonomous workspace that keeps artificial intelligence coding agents running all day and all night. The company said it developed Crew to allow developers to start a task, walk away and come back when something is worth administering. It tasks multiple agents and their sub-agents on a multi-step task so developers can let them run unattended. It can also manage recurring jobs that run on a schedule, and a heartbeat monitors the system until something needs to be reviewed. The software also provides Apps, an interface that allows developers to bring their own tooling, giving individual processes their own purpose-built UI and structure. An App can combine a custom user interface with agents, skills, schedules and backend services. The new solution is built on Kiro, expanding on the same agent infrastructure of the software’s autonomous mode. Kiro is AWS’s development environment that integrates agentic AI for software engineers, which allows them to rapidly turn ideas into production-ready code. The difference between Kiro Crew and Kiro’s “autonomous mode” is one of scope: autonomous mode handles individual tasks within a single session; Crew continues working whether a developer is online or off, maintains memory across sessions, runs scheduled work and orchestrates multiple specialist agents. As this entire system is built on Kiro, it can read .kiro configuration out of the box, built off steering files, skills and custom agents the user has already built as part of their coding environment. One thing that Amazon touted about Crew is that it is designed to be “self-learning and self-evolving.” By providing persistent memory of preferences, active-project context, and relevant history into new sessions, it can bring corrections along with every run. This includes lessons that change later behavior involving workspace-scoped or project-specific guidance. For example, if a user repeatedly asked an agent to code in a particular way, such as a preferential use of a certain function over another, the agent would ingrain that guidance. Also, if a specific tool, for example an MCP server that provides Python best practices, constantly fails on certain requests that need to be reformatted to get it to work properly, the agents can pass on this information between sessions. Amazon stressed that agentic memory remains visible, allowing developers to decide what Crew carries forward. Interacting with and expanding Crew Crew runs locally or remotely on developer hardware, providing the ability to view, audit and control. The app and web dashboard provide a place to work directly with Crew across conversations, files, approvals, memory, schedules and Apps. Developers can also talk to Crew via their usual chat interfaces. This means reaching out to the agents via Slack, Telegram and Discord, allowing them to continue the same work without needing to stay in the same place. This means that a developer could start their tasks in the Crew dashboard, then walk away. If they’re off doing something else and realize they need to add a task or adjust something, they can contact Crew via Discord, or have an agent check in when something needs review. The core of Crew is also extensible, allowing the community to build an ecosystem of Apps and integrations that expand what it can do. The company said this community-driven capability became the core of Crew when it first emerged among Amazon’s own developers and going forward this will be preserved. For developers