AWS Kiro Crew Ships Bots GA: Named Long-Running Agents with Own Memory and Parallel Sessions — Delegate Processes, Not Tasks

Core Background & Industry Pain Points

Parallel coding agents turn developers into dispatchers tracking sessions and reconnecting context. On November 5, AWS open-source environment Kiro Crew generally available Kiro bots so the unit of delegation becomes ongoing process ownership, not one-off tasks.

Architecture Highlights & Internals

Each bot has a name, avatar, persistent Notes memory and its own workspace; one ongoing chat sets direction and returns results. A bot can start worker sessions in parallel, monitor them, redirect drift, and consolidate outcomes; when a human decision is needed it asks while continuing independent work. Schedules can be set in chat. Tools are minimized per role; bot and workers can use different models. Dashboards are described in natural language and maintained by the bot.

Authoritative Benchmarks & Measured Scores

No independent agent benchmark scores shipped with the launch; this is an orchestration and process layer. Third-party coverage (InfoWorld, VentureBeat) highlights lower parallel-agent coordination burden while flagging token and compute cost plus audit and governance risk from long-lived memory. Do not invent scores.

Developer Hands-on Guide

Open the bots workspace in the Kiro Crew sidebar, choose Add a bot, and base it on the default agent or an existing custom agent to reuse prompt, model, tools and skills. Start with one reviewable recurring responsibility such as issue triage or CI health, set boundaries and escalation rules, then grow the crew. See kiro.dev/blog/crew-bots. Audit Notes over time; do not habitually approve every tool-expansion request.