Migration guide for AWS developers as Amazon Q reaches end-of-support in favor of the new agent-centric Kiro reasoning platform.
Why Amazon Q Is Being Replaced
Amazon Q has served as a coding and cloud assistant inside the AWS ecosystem, but AWS is shifting that role to the Kiro reasoning platform. The change is not a simple rebrand. Amazon Q was built around question-and-answer and inline help. Kiro is agent-centric: it plans multi-step work, calls tools, and carries context across longer tasks. That design matches how many teams already use AI in development—not as a chat box, but as a collaborator that drafts, checks, and iterates.
End-of-support for Amazon Q means existing integrations, shortcuts, and team habits will stop being the supported path. Treat the transition as a product migration: inventory what you rely on today, map it to Kiro’s agent model, and retire Amazon Q-specific workflows before support ends rather than under pressure later.
What Changes for Day-to-Day Development
Under Amazon Q, a typical flow was ask a question, get an answer, paste or apply the suggestion. With Kiro, you define a goal, constraints, and the tools the agent may use—repositories, cloud APIs, runbooks, or internal docs—then review and approve the steps it proposes. The unit of work becomes a task with state, not a single prompt. That improves multi-file refactors, infra changes, and cross-service debugging, but it also raises the cost of vague instructions and weak guardrails.
Expect more emphasis on permissions, auditability, and human checkpoints. Agent runs that touch production-adjacent systems need the same review culture you apply to infrastructure-as-code PRs: least privilege, clear success criteria, and a path to stop or roll back mid-run.
A Practical Migration Path
Start with a short inventory. List where Amazon Q is used today: IDE plugins, console helpers, chat in tickets, CI annotations, or shared team prompts. For each use case, note inputs (code, logs, account context), outputs (patches, commands, explanations), and who must approve before anything is applied. Then map each item to an equivalent Kiro agent setup: system prompt or policy, allowed tools, and review rules.
- Pilot on non-critical repos and sandbox accounts first.
- Rewrite team prompt libraries as agent goals plus constraints, not one-off Q&A snippets.
- Replace “ask Q, then run the command” habits with agent plans that produce a reviewable diff or change set.
- Update onboarding docs, runbooks, and internal wiki links that still point at Amazon Q.
- Decide what stays human-only (secrets handling, production deploys, cost-sensitive scale-ups) and encode that as hard limits for agents.
Keep Amazon Q available only for comparison during the pilot. When a Kiro workflow matches quality and latency for a given task type, switch that task type fully and remove the old path so teams do not maintain two mental models.
Risks, Validation, and How to Stay Productive
The main risks are incomplete tool access (agents that cannot see the systems they need), over-broad access (agents that can change more than they should), and silent quality drift when prompts are copied without success metrics. Define a small set of golden tasks—common refactors, log diagnosis, and a safe infra change—and run them the same way before and after the switch. Judge outcomes on correctness, review effort, and how often a human must intervene mid-task.
Document ownership: who maintains agent policies, who can grant new tools, and how failures are reported. Treat Kiro configuration like application code—versioned, reviewed, and tested—rather than as personal chat history. That discipline turns the Amazon Q end-of-support window into a controlled cutover instead of a scramble when the old assistant is no longer available.