I don't need an autonomous agent - I need a team mate that grows with me
Something new from the Kiro team
I’m fortunate. Being an AWS DevTools Hero gave me the chance to experiment with the Kiro autonomous agent very early. I got access to the agent the day it got announced, and since then I’ve used it (extensively).
I’ve shipped over 35 features, bugfixes, enhancements for Nexus Share and I’ve also used it to progress on a few other projects. The first “feature” it shipped, directly during re:Invent 2025, was an integration of Nexus Share with my URL Shortener “lckhd.eu” (there is no UI for that).
The longer I work with the agent, the more it becomes apparent to me that “this type of agent” is not what I need.
Let me try to explain what I mean by that in telling you some more details on my working habbits and my thought processes.
Making progress with my products, tools and features
As some of you might have seen, I’ve not been very active on YouTube in the last few months. Instead, I launched Nexus Share which I invite you to try out, I created a relevant YouTube channel and I’ve become more active on X.
Overall, I’ve been playing around with Agentic AI assistant coding using Amazon Q developer and, since I can use Kiro with my Identity Center (see blog) also with Kiro.
Nexus Share is about 85% “Vibe coded” - this means that the different coding agents have contributed the code to it. I have mainly used the IDE - VSCode or Kiro - to create the code as it gives me the chance to directly interact with them.
The tech stack is TypeScript (which I know… roughly…) and Flutter (which I’ve used but I don’t know a lot about). What I have built would not have been possible without Q Developer and Kiro.
But, as you know, working with an IDE requires
- consistent laptop / desktop to be available and running (and is limited by CPU/GPU/RAM)
- consistent hand-holding of the “coding assistant”
…and is currently also limited to working on “one” branch and repository (unless you get very creative…).
The Kiro autonomous agent (and alternatives like Ona or extremes like Lovable.. or the other copycats of those) promise to solve a lot of the problems with “local” IDEs and partly promise to re-invent software development.
I was really curious to look at what the Kiro autonomous agent is able to do - so let’s look at it closer.
Working with the agent
Kiro autonomous agents - An additional step into Agentic Coding support on AWS
At re:Invent 2025 AWS announced the availability of three different “frontier agents”: DevOps and Security Agents and, last but not least, the “Kiro autonomous agents”. In this blog post we are goig to look at how the Kiro autonomous agent works and how this can help you to expand “agentic AI” beyond your own local device easily.
The idea and main entrypoint
The idea of the Kiro autonomous agent is to make the power of Kiro accessible in a simple way and execute coding tasks without you supervising the AI agents the whole time. Under the hood, the agent seems to be using the same technology and models that the Kiro IDE uses.
The idea of the autonomous agent is that give users the possibility to interact with the agents through a simple web inteface, providing access to a simplified chat interface and integrating directly with your code provider. For our test, we have used the Github integration.
Accessing the agents
You access the agents through kiro.dev or specifically app.kiro.dev. After connecting the app to your code repository (e.g. Github), the agent gets access to the repositories you have selected. Now you can start to interact with the agent!
Once you create a new chat, the agent starts working on your task. That means, the agent will initially start to clone your repository (or repositories, more on that later!) and will analyze your code base - before starting to work on the task in the same way the Kiro IDE integration would do that.
You will now be able to see the agent working on your task with regular updates being given in the UI. But you could also do something else - what I tried out at re:Invent: turn off the PC, go to meet some friends, have fun and enjoy some time without being stuck on the PC!
The agent will be working on your task(s) even if your session ends.
What the Kiro autonomous agent can do for you
Right now, the Kiro autonomous agent can work on one or more of your connected repositories and can help you analyze your code or work on tasks. It does currently not support the approach of “spec driven” development built into the interface, but the agent behind the chat interface is able to do the same things that the agent in the Kiro IDE is doing.
The agent can also implement code for you and it contributes it to your repository or repositories by raising Pull Requests against the “main” branch.
You, the “human in the loop” can then review the Pull Request and you can interact with Kiro either through the Pull Request (by triggering the agent to review your feedback with a comment) or you can steer it again in the Kiro Web UI.
The capabilities overall are good.
I’ve not yet been able to verify the “memory” feature of the Kiro autonomous agent - one of the bigger innovations where the agent will keep a history of “you” as a developer accross multiple sessions.
Learnings so far: What the Kiro agent - in the current materialization - misses
The Kiro autonomous agent currently feels like a handcuffed exert. The (simple but usable) UI is a chat interface and the Github intergration seems to be built on top of the Q Developer in Github functionality.
Initially, it was not clear to me wether I would be interacting - after the agent finished - through the Kiro Web interface or through GitHub… I learned that I need to take that decision myself right now.
There is no integration between your Kiro IDE and the autonomous agent. This is something that is very limiting to me. I would have expected to be able to “seemlessly” switch between my IDE and the autonomous agent - to allow me to quickly “jump” into the current state of code that the agent is working on - but that’s not available yet. And I am not even sure if that’s what I would love to get in the future!
Please don’t forget, the Kiro autonomous Agent is in PREVIEW and I’m expecting a lot from the team - as a Preview, this is a very well executed, initial release!
What I would like to get instead and why I don’t think an autonomous agent solves my problems
The Kiro autonomous agent - at least for me - has not become a team partner that grows with me. What I am looking for is one or more “autonomous agents” that become team mates and partners for me, learn from my (or better “our”) mistakes and get better every day. I would like to be able to ideate with a partner, that gives me honest feedback and asks questions. My “coding assistant” should be able to decide autonomously when to build something and when to continue to brainstorm.
This assistant should be accessible everywhere: web, mobile, desktop. And the transition should be seemless, as I’d like to continue my thought process(es) while being outside or waiting for a doctor appointment.
Besides all of that, I do also believe that the “human in the loop” is important - and that means that I would like to - at any stage - step into the current state of the code to be able to steer the agent better. Essentially, I also would like to be able to tweak the code at any time.
And this means that I need to be able to step into the “Kiro IDE” at any stage of the process.
Let’s get back to the “partner” idea - think of my “coding agent” as a partner with personality that matures with me and along the way - it needs to get better and learn from mistakes and then introduce that knowledge for our next conversations. And then, potentially also get to know better what my intent looks like.
And then, in the future, ideas might be created by “my personal coding assistant”?
Will we see this kind of assistant in 2026?
No. 2026 will be a year of exprimentation and maybe, in the second half, consolidation. There is too many different “AI assist” coding tools out there that all try to do the same thing - and under the hood often use the same AI models.
Have you seen something close to this somewhere?__
Please let mek now - send me lockhead+blog@lockhead.info an email or reach out to me on one of my socials!
Are you playing around with the Kiro autonomous agent? - Drop me a message aswell, I’d love to talk!