Agentic AI
AWS Kiro Bots Give AI Agents Their Own Memory, Freeing Developers from Session Babysitting
AWS is adding persistent agent memory and workflow management to Kiro, letting developers delegate entire processes rather than manually shepherding tasks between sessions.
1/3
The New Stack
AWS has expanded its Kiro agentic development platform with Kiro bots—long-running agents that retain their own memory and manage parallel workflows independently, reducing the need for developers to manually track and connect agent sessions.
Launched on Thursday, Kiro bots build on Kiro Crew, AWS's open-source workspace introduced in August. While Crew preserved context across developer sessions, bots give each individual agent its own durable memory store and the ability to spin up and oversee multiple parallel workstreams without human intervention.
From Task Delegation to Process Delegation#
The core problem Kiro bots address is what developers call context rot—agents lose continuity between sessions, forcing engineers to re-explain decisions and re-feed information repeatedly. Kathy Kam, AWS's director of engineering for Agentic AI, explained the shift: "Developers can now spend the day deciding what's worth building, instead of babysitting sessions."
Each bot operates with scoped tools and permissions tailored to its specific role. A research bot might search and read files; a deployment bot can execute scripts and request approval before changes. Bots store their learnings in markdown notes that developers can directly edit, fixing errors once rather than watching them resurface across multiple parallel tasks.
Trust Through Scope, Not IQ#
Kam pushes back on the idea that bots form some kind of agent hierarchy. Instead, trust is earned through restricted authority and track record. "A hierarchy of agents based purely on IQ is the wrong measure. What matters is what an agent is allowed to touch, and a Kiro bot earns trust by building a track record."
Each bot has a dedicated chat channel where developers set direction and collect results. Behind the scenes, the bot launches parallel sessions, monitors drift from the plan, and can ask for human decisions on blocked tasks while continuing work on independent ones. Teams can establish routines so bots investigate overnight failures before engineers arrive.
Keeping Humans in Control#
A persistent tension remains: as agents accumulate institutional knowledge, when should human judgment be bypassed? AWS maintains that developers remain accountable decision-makers. Bots can request additional authority but cannot grant it to themselves. The human sets boundaries for each bot's domain—how much approval is required, which tools are accessible, where judgment belongs.

