Kiro Crew: the productivity tool that boosts my agentic AI

A bit of history: from Q CLI to Kiro CLI

• Develop (application or infrastructure code)
• Diagnose operational issues (it excels at calling AWS APIs, such as CloudWatch, RDS Data API, RDS Performance Insights, etc.)
• Handle FinOps tasks
• Write blog posts covering technical projects
• Contribute to open-source solutions
• Update my website (specifically importing blog posts from dev.to)
• Assist with responding to RFPs (Requests for Proposals
• Write scripts and demo setups for my training courses.

• I quickly gave up on using MCP servers; the essentials are covered by native tools or the terminal. The only MCP I’ve used in the last six months is the Figma MCP.
• I generally keep Kiro on a tight leash , which yields excellent results with very low credit/token consumption (especially since I always use "auto-model" mode, which selects the model best suited to the task's complexity); I still don't understand how people manage to burn through a month's worth of Claude subscription in just two hours.
• I don't use sub-agents (partly because managing them via the Kiro CLI is often clunky—especially regarding permissions—and partly because, despite the Kiro team's genuine investment in the TUI, managing multiple agents in parallel isn't practical). Ultimately, my multi-agent workflow relied more on simply having multiple terminal instances running.
• I don't use Kiro's memory feature; instead, I tend to have very long conversations per project (here is my most-used terminal alias: alias kr='kiro-cli chat --resume-picker --tui'), as Kiro CLI partitions conversation history based on the directory from which it was launched. The conversation history itself serves as the memory (history compaction has improved significantly in recent months, so I no longer experience major data loss when it occurs). Some of my Kiro discussions have histories spanning 40,000 messages.
• I hardly ever use custom agents (except for website-related tasks, where a detailed prompt allows the tool to operate 100% autonomously using just a blog post URL). Not "skills"—simply because Kiro didn't support them until January, and I missed the announcement that support had been added 😅—nor hooks. Consequently, despite having Makefiles and Markdown-based READMEs/steering files for each project, my agents tended to reinvent the wheel and run "inline" Python scripts for complex tasks.

What is Kiro Crew, and what has it changed for me?

• As I understand it, Kiro Crew includes:
• a graphical interface (the "dashboard") that provides an easy overview of all active sessions (including sub-agents), task scheduling, and settings management (specifically for configuring custom agents and skills);
• the dashboard allows you to connect to one or more "gateways"—the machines where the agents actually run. This means you can execute tasks locally or on a remote machine... I have a feeling I’ll soon set up a small VM so my agents can keep working while I’m traveling or when my machine lacks a reliable internet connection;
• gateways can
• communicate via various channels: the dashboard is the obvious primary channel, but you can also connect Slack, Teams, Discord, etc. and thus manage operations remotely or as a team.
• manage scheduled tasks (cron); for instance, I can launch a weekly FinOps review. Agents can interact with the gateway to schedule tasks.
• run custom applications (with a marketplace that will be rapidly expanding).
• a shared knowledge base (automatically kept up to date by agents based on the directives we provide).
• integrated skills (such as Apple's "Computer Use" or controlling Chrome via the Playwright extension).
• voice control (using either 100% local text-to-speech/speech-to-text solutions or AWS services). Full disclosure: I haven't used this yet, as I’ve been on vacation and mostly experimented during nap times or in the evening, in environments where I couldn't make noise.

• define an agent capable of preparing a response to a call for tenders; the orchestration of scheduled tasks and Playwright allowed me to tell the agent: "Check in three days if the consultant I contacted via LinkedIn has agreed to my co-contracting proposal, and if so, add their commercial references to the bid."
• remotely diagnose a DDoS attack on a client's infrastructure while at the beach 🏖️, gathering a set of facts backed by Log Insights queries and recommended mitigation actions (which were implemented that same evening).