Going Quickly with Quick and Kiro

I've been finding myself spending less and less time using Kiro, at least directly. Since it launched headless mode and ACP I started using my own custom UI  over the top of it. I get all the benefits of Kiro, it's vast model support, steering and hooks  I've already made, but in a UI that worked for my specific workflow. But, I don't really want to be endlessly building features into a custom tool for me. I'll never build new features as fast as a team dedicated do it.

Quick Desktop launched two months ago and since then it has become my default UI for almost everything I do on my laptop. Whilst Quick can write code, it can also orchestrate multiple coding assistants for you. It has built in support for Kiro, Claude, Cursor and Codex. But because you can configure your own commands for how it launches a coding agent, you can take full advantage of Kiro's custom agents and model library. And this unlocks some very powerful use cases.

I have a custom Kiro Agent for doing research and planning. It has the model and the tools I want, with the trust boundaries I need. A custom agent with the model I want and the permissions I need for it to write code. And a custom agent with another model for reviewing finished changes. Quick handles the outer agentic loop for me. Working with me to find work, delegate to coders to do it and managing reviewers to check it. For me, it's becoming an out of the box management system for the full AI-SDLC.

Quick has many built in Connectors that use both MCP and API integrations to various SaaS services and allows you to add your own. Quick also allows you to configure your own custom MCP connections, allowing you to bring your own client credentials or using DCR in the oauth 2.1 spec. An LLM or even an agent by itself isn't overly useful unless it has access to a wide range of tools, and Quick supports everything you could need.

Finally, Quick self learns over time. It builds its own memory of useful things and builds a knowledge graph of connected entities that you work with regularly. This could be the people on your team, your peers, a Slack channel you work in, a Github repo you regularly commit to or a pattern you use when working. Quick self categorises your working life and uses it when needing to take action or give you guidance. Once you've connected Quick to your email and Slack it learns from your conversations and knows what you need to do.

When you pull this together, it's easy to see why Quick is my default application for my day to day work. Every day it reads my Granola transcripts and keeps its knowledge base up to date. Right before meetings it gives me a summary and background to the discussion. It reminds me if I've forgotten to answer a Slack question and summarises the important bits of discussions for me. Sure, it might miss something, or get it wrong. But I do that too. It just works better than I have been able to in the past. Whilst these help with some admin, planning and decision making parts of my role, a large part of my role is also feature output. I call it feature output, because coding is only part of it and I don't do the coding.

Because Quick is already ingesting a large amount of auxiliary data around the applications I work on, and knows of the decisions I've made, it's able to orchestrate the coding parts. Every couple of days I get Quick to help manage my backlog and build some tickets based on issues it has seen from Slack feedback, my own feedback, error logs and bug reports. It puts together a summary. If it needs some help understanding the code, it knows where to read and will delegate to Kiro to inspect the bug and Kiro already has all my steering configuration. Quick knows how I want my code flows to work. I tell it what ticket to work on, it delegates to a Coding Kiro to do the work. When that Kiro thinks it has finished, Quick delegates it to a Reviewer Kiro to review the code. If it finds issues then Quick orchestrates across the coding instances for me until the reviewer gives it a LGTM. It directs the software coders for me. It manages the loop. It knows what's going on.

Quick's abilities to manage loops, schedule tasks, give me just in time information and call in custom agents exactly when it needs to is changing the way I work, and I don't think we're going back. The value add I find myself bringing isn't writing code. It's all the bits of software engineering around it. Quick amplifies my ability to do this. It does the busy work for me. I don't write Jira or Linear tickets any more. I review them when I need to. I don't write code, I review the team who are writing the code. I don't spend time finding documents or managing meeting notes, I get given exactly what I need, exactly when I need it.
This is a crowded space. But Quick’s automatic knowledge bases, self improving knowledge graphs, memory and seamless integration with a wide variety of coding agents has allowed me to drop any custom tooling and it fits right into how I work. Because of all its capabilities, I think it can fit into how almost everyone works today. This isn't a software engineers tool, it's a tool for people that work on products and it's proving to me to be a great tool for me. 

Quick does more so I can do more.