Plus, four features of Kiro Crew that unlocked productivity I never thought I'd have.
I'm testing AWS services on a weekly basis, so I use agentic coding tools to stand up AWS apps all the time. I had one thing on my "AI developer tools capabilities from AWS" wishlist... intuitive control over my agents. I wanted to let my agents loose to build anything I tell them, while retaining control over which tools and shell commands they execute.
It seems that my wish was granted by the team behind Kiro, the agentic coding assistant built by AWS. They recently released Kiro Crew, an open-source agentic development platform that augments Kiro with new functions while retaining its most helpful features. Let me show you some of my favorite features of Kiro Crew that have more than doubled my development speed.
But first... What is Kiro Crew?
Kiro Crew is a workspace optimized for managing multiple complex agent workloads. Think of it as an augmented interface for the Kiro agent infrastructure. You're using the same underlying Kiro agent, but with an expanded agent harness that brings more flexible chats, pre-built apps, and other features that improve your ability to interact with the agent you already use.
I've been trying every which way to build code in Kiro Crew, and I've already implemented a few of its features into my daily work, including custom agent profiles, CRON jobs, and even just the new UI layout. One feature has completely changed the game for me, though, and that's Kiro Crew's Channels feature.
1. Kiro Crew Interfaces (Slack, Telegram, Discord, etc.)
Interfaces (referred to as "Channels" in some docs) allow you to connect Kiro Crew to platforms like Slack or Discord, primarily through the Kiro Crew Gateway. For example, I created a Slack app on my company's Slack environment through which I can control Kiro Crew when its active on my laptop. This includes starting and communicating with chat sessions (tagging it with @kirocrew and sending a prompt), configuring settings (/kirocrew config), toggling YOLO mode (/kirocrew yolo, auto-approve all commands for 6 hours), and even switching between Kiro Crew Agent profiles (/kirocrew agent, which lets you switch between pre-configured agents that are specialized for different tasks).
Monitor Time-Consuming Jobs Away from My Desk
I'm no saint when it comes to discipline with my agents, as I love to let them loose on my codebase as much as the next guy, but I strictly prohibit my agents from running destructive commands without my approval. As such, when building complex apps for internal workflows, workshops for our Student Builder Groups, et cetera, I often get several tool requests every minute. I review the reasoning behind every shell command, and while it leads to a better overall product, I know that my workflow will get hung up on an approval prompt within a minute of me leaving my computer.
Now, with the Slack channel configured, I get approval requests sent straight to my phone through the connection I set up between Slack and my instance of Kiro Crew's gateway.
In this example, I ask Kiro Crew to review my recent projects, create a markdown file listing open issues in my code, and save it as a to-do list in my Projects directory.
As it builds, it will stop it's own turn if it encounters a design decision that would impact how it builds the project going forward, like in this example.
Now, I'm fully set up to continue my projects from anywhere, without splitting my committed code between devices. Follow these steps to set up a connection to your preferred platform.
2. Kiro Crew Task Runner
Task Runner is a built-in App designed for orchestrating massive multi-agent tasks. If you've used the Kiro IDE, you can think of Task Runner as a similar workflow. Task Runner will ask you to "describe your task". You can either click "Run" right out of the gate (similar to "Vibe Mode" in the Kiro IDE), or you can click "Refine into Spec" to have Kiro organize your prompt into an actionable task list for Task Runner to design the workflow against (similar to the Kiro IDE's "Spec mode" workflow).
You can also select which Agent profile will handle your task, which lets you select a model to use.
The initial task list was broad (unsurprising, given the one-sentence prompt I used to begin with), so I had a separate agent use Sonnet 5 to create a task list to feed to Task Runner.
Once you direct it to execute your task, it orchestrates subagents to tackle the task list in parallel. Each agent retains its own context window and focuses on a specific subtask, improving each individual agent's performance and reducing the likelihood of hallucinations compared to dumping everything into one agent.
Now, I can let this agent run in the background as I work on other tasks, or walk away from my computer. *NOTE: If you do not enable "YOLO" mode, Kiro Crew will still ask you for your approval to execute commands if commands are not pre-approved.
An hour or so later, with no interruptions, I returned to a completed chunk of development work. A quick playtest confirms the changes I made were implemented.
3. Kiro Crew Scheduled Tasks
Kiro Crew lets you run pre-structured agentic workflows on your device. By default, Kiro will stagger your LLM-backed CRON jobs in order to not throttle your computer's resources, but you can turn this off as a setting. You can select an agent / model, write the recurring prompt, and even select a channel within an Interface (slack / discord, etc) to send your report to.
Here's an example of a scheduled job I've set up on my machine.
Here's the full prompt:
"Every morning at 9:00am EST, generate and send a report in slack detailing the following:
- Memory constraints on my device (and how constrained Kiro Crew is due to memory)
- Active tasks being worked on by agents
- Design gaps and unfinished work that either needs to be completed or is blocked requiring a design / architecture decision from me, the developer.
This report should be organized into clear sections, have concise action items, and only include relevant information I need to get up to speed. I will ask clarifying questions if need be, so err on the side of conciseness. NO FILES SHOULD BE CHANGED WHATSOEVER ASIDE FROM A REPORT IN A MESSAGE SENT VIA THE SLACK INTERFACE."
After test running the report, here's the result, right in my Slack environment so I can get notified and read it from my phone:
NOTE: The Scheduler will NOT retry jobs if the gateway is offline at the scheduled time for the job. You may have to re-run a job if your connection to the gateway is disconnected.
4. Persistent Preference Memory
Kiro Crew learns how you work. It analyzes your development habits as you use the platform, quantifies your development philosophy, and adheres to your preferences as you build projects.
How It Works
Every interaction you have with Kiro Crew feeds into a persistent memory layer. When you reject a suggestion, approve a command, rewrite generated code, or configure a setting, Kiro Crew logs that decision. Over time, these observations compound into a development profile that influences how Kiro approaches your projects.
For example, when I direct an agent to execute multiple tasks, I tend to ask it for a summary of changes along with several passed tests for each artifact it builds. After just a week or two of using it, this is now a default behavior my instance of Kiro Crew employs in response to my multi-task prompts.
Sessions: Context That Persists
Kiro Crew organizes work into Sessions, which are persistent conversation threads that maintain full context across days or weeks. Unlike a typical chat window that resets when you close it, Kiro Crew Sessions retain:
- The full conversation history and decisions made
- Files created, modified, or referenced
- Tool approvals and rejections you've issued
- Architectural decisions and their rationale
This means you can close your laptop on Friday, reopen a Session on Monday, and pick up exactly where you left off. Kiro Crew remembers what you were building, what constraints you established, and what approaches you already tried and rejected.
Memory in Practice
I have a few preferences that I've never explicitly documented anywhere:
- I prefer functional React components with hooks over class components
- I want error handling to be explicit in most of my MVPs, to make it easier for me to debug when playtesting
- I deploy to
us-east-1unless there's a latency reason not to
After about a week of working in Kiro Crew, these preferences were internalized. When I spin up a new project, the scaffolding already reflects my conventions. When I ask for error handling, it writes explicit try/except blocks with logging rather than bare except: pass patterns.
Managing Your Memory
You're not locked into learned behaviors, however. Kiro Crew exposes your preference memory through the settings panel, where you can:
- View all learned preferences with confidence scores
- Edit or delete specific memories
- Pin preferences you want enforced regardless of context
- Scope memories to specific projects
This is the feature that compounds over time. The longer you work with Kiro Crew, the less friction there is between your intent and the output. It's the difference between dumping thousands of tokens of context every session just to catch your agent up to speed, versus maintaining a seamless flow between session refreshes.
What to take away
Kiro Crew turned me from a basic AI user to an agentic puppet master, and my agents feel more efficient than ever... now, I encourage you to try integrating Kiro Crew into your workflow the same way. By consolidating all of your sessions, enabling control of Kiro from your phone, and waking up to automatic performance reports and to-do lists, Kiro Crew helps maximize your productivity across complex projects.
Give these features a try yourself, starting at the installation page!
Top comments (1)
Just what I needed to get started with Kiro Crew! Thanks!!