Go From Vibe to Viable With Kiro, the Agentic IDE for Real Software Development
AWS launched Kiro early last week, a standalone “agentic” IDE that directly integrates LLMs into the developer’s code editor to speed up and simplify development. Kiro is built on Code OSS, so you can use your existing VS Code settings and Open VSX compatible plugins. It is currently free to try during its current preview period; due to high demand AWS added a waitlist for access.
What sets Kiro aside from competitors in this space is the focus on an interface and tool set that makes it easy to deliver on good software development practices, including requirements documentation, design specifications, and task planning.
Early feedback on Kiro has been positive, with users drawing favorable comparisons to other AI coding tools like Cursor, Windsurf, and Lovable. This is a market that has a lot of attention right now (see, for example, Google’s mid-July $2.4B acquisition of Windsurf’s engineering team out from under OpenAI’s nose), and rising above the noise is no small feat.
Before we talk about what Kiro does differently, let’s talk a bit more about the competition.
The Market Today
All coding tools have some element of human-in-the-loop, but they differ in how they design the loop. For example: What types of decisions can the tools make on behalf of the user? How often and in what way does it return results to the user for validation?
In the smallest loop, tools will provide suggestions to the user on the code they’re working on – streamlined function completion or chat windows that offer suggestions for debugging. Think of some combination of autocomplete and the old Clippy Office assistant. While these features can be useful, they’re not much of an improvement on existing tools. Code completion has been around for decades, and we all have thoughts about Clippy.
The broadest loop is probably vibe coding, a recently minted term for generating code using only natural language via LLMs. Vibe coding tools like Lovable and Bolt greet you on their front page with a text box and encourage you to “create stunning apps & websites by chatting with AI.” These tools behave in ways that are much more “agentic,” able to build and execute a plan and adapt to meet a user’s goals. Rather than authoring code, the user becomes the conductor, orchestrating the development of their idea through the natural language interface. Perhaps counterintuitively, this might not actually speed up development, particular for experienced developers: one study from METR found that experienced developers worked slower with help from AI tools, despite those developers’ own assumptions to the contrary.
The front page at Bolt encourages you to create your next website by chatting with AI.
These tools promise to open up the power of coding to everyone. According to the media, it’s an approach that some businesses are starting to explore. But as you might expect, this approach is risky and error-prone, particularly for larger projects, and it can lead to unfortunate outcomes like making your entire product waitlist publicly visible on your website’s front end, or deleting a prod database despite user instructions not to do so.
From Vibing to … Viable-ing?
Kiro’s core difference comes back to this insight: Experienced developers working on complex enterprise production systems don’t need or want a machine that will only write lines of code for them; they need help with the other stuff, which can enhance their ability to code well. The majority of developers’ time is not spent writing production code. It’s spent on other stuff: learning about and maintaining existing code, writing documentation, troubleshooting issues, drafting designs, thinking through problems, documenting specs, aligning with product teams, building tests – and the meetings, chats, and emails that go along with that work. This is particularly true for senior engineers, whose role is often describing the work they want done to less experienced staff rather than doing the hands-on coding themselves.
In other words, the goal is to simplify and streamline non-coding work in ways that give expert developers the ability to better apply their experience, knowledge, judgment, and taste to the code they deliver, in a way that multiplies the value they can offer.
Kiro helps in two main ways: Spec Mode and Hooks.
Spec Mode
LLMs struggle to deliver useful results when instructions are vague or when the context is too broad. Spec Mode helps solve this problem by translating high-level ideas into technical specifications, detailing what should be built and how it will behave. This helps the developer create a plan of attack for the project and helps to align what they’re doing with the rest of the design, development and product team.
Kiro will take your requirements in EARS (Easy Approach to Requirements Syntax) notation (or generate them for you from natural language prompts). In a second tab, the tool turns those requirements into system designs, including architectural components and data flows. A third tab creates implementation tasks, which you can have the tool execute one at a time. At any point in the process, you can directly modify any of the outputs, or you can direct changes in natural language via the chat window.
Hooks
Hooks operate like lane-assist for coders: automatic triggers that execute predefined or agentic actions to keep a project in line with the organization’s standards. Hooks can check a project for security vulnerabilities, run linting or formatting, run tests, identify deviations from documented standards, and create documentation.
Like any other autonomous workflow, you must ensure you set very clear objectives and target metrics for the hook, test before you deploy it and set appropriate guardrails, and measure the difference between the applied hook and your target metrics. Then, take action if you start to see significant drift from the target metrics.
And You Can Still Vibe if You Want to
Not all work requires the rigor of spec coding. If you want to explore or test something out or implement a quick task, Kiro still has a Vibe Mode that allows you to go straight from abstract ideas in natural language into code.
Our Take: Do This Now
- Gain hands-on experience with AI-enabled development tools now. If you haven’t looked into AI-enabled coding assistants and IDEs and you conduct any amount of internal development, you need to start getting hands-on experience with these tools today. Most tools offer a free preview and a ~$20 monthly subscription after that, so the upfront cost of a limited trial can fit within a discretionary or training budget for most organizations. Kiro is based on VS Code, so the look and feel of the interface will be familiar to many of your developers already. Hands-on experience with this type of tooling is a prerequisite for any IT organization that wants speed up the delivery lifecycle, as you might with Boost Solution Delivery Throughput With AI, to understand how and where in the SDLC these tools can offer the most value.
- Use our research to launch your scaled-out proof of concept. In particular, use our Proof of Concept Workbook to streamline the key decisions you need to make to scope, evaluate, and plan a larger implementation.
- The real magic happens in an environment where developers can explore and share new tools and techniques. This is about people changing the way people work, and people won’t change if they don’t think there’s value in it. At AWS there were inflection points, sometimes traceable to a single event or post. Erin Kraemer, an Executive Technical Product Leader at AWS Agentic AI, told us internal adoption of coding tools happened when developers heard about “someone they trust who’s doing something they find interesting and trustworthy” and decided to try it out for themselves.
The Bottom Line
We haven’t seen innovation in IDEs like this in years. Software development leaders should be actively investigating coding assistants that simplify and streamline non-coding work, to enable your experienced developers to deliver robust, secure, and serviceable code.