From prompts to specs: AWS’s Kiro signals the next phase of AI coding tools

AWS has previewed Kiro, a new agentic AI-driven integrated development environment (IDE) that the hyperscaler claims is better suited for structured, large-scale software development workflows, favored mostly by enterprise engineering teams seeking automation, traceability, and production-grade code quality.

Unlike traditional AI coding assistants that rely on ad-hoc prompts, Kiro introduces a spec-first workflow that mirrors how large engineering teams plan, design, and ship code, making it a strategic choice for enterprises prioritizing reliability, documentation, and collaboration, said Jason Andersen, principal analyst at Moor Insights & Strategy.

Spec-driven development (SSD), in the context of agentic IDEs, such as Kiro, means a three-step process where a developer or a team of developers first defines what the application should do, and then an agent or agents build and test it to match the specification defined by the team.The third step, often described as the refinement phase, involves revisiting the initial specification to reflect any changes or insights gained during development, ensuring it accurately represents the final application.

How does Kiro work?

Developers or teams start by telling Kiro what they want to build, and instead of jumping straight into code, Kiro turns the idea into a clear, step-by-step plan using EARS (Easy Approach to Requirements Syntax).

It breaks down what needs to happen, including different situations the system should handle and how to know when it’s working correctly.Kiro then analyzes the team or its enterprise’s codebase and the approved specs to generate a full technical design. This includes data flow diagrams, TypeScript interfaces, database schemas, and API endpoints. It eliminates ambiguity and aligns the team on how the system should behave.

Based on the design, Kiro breaks down the work into sequenced tasks and sub-tasks, each linked to the original specified idea. These tasks include everything from unit and integration tests to accessibility and responsiveness checks.

Developers can trigger and monitor these tasks step-by-step, with full visibility into code differentiations (diffs) and agent actions.

Kiro also comes with agent hooks — event-driven automations that run in the background when files are saved, created, or modified.

These hooks handle repetitive tasks like updating tests, refreshing documentation, or enforcing coding standards, acting like a vigilant teammate who never misses a detail, the company explained in a blog post.

Compared to other AI coding tools, Hooks is a clear step up in control and extensibility, said Derek Ashmore, AI enablement principal at Asperitas Consulting.

However, he warned that it’s early days and said that while hooks add real utility, they are not a game-changing leap — yet.

Rather than mainstream developers looking for a drop-in Copilot alternative, hooks as a concept is likely to appeal most to power users, tool-smiths, and startups building on top of Kiro.

How does it help developers?

Analysts pointed out that Kiro’s spec-driven development approach could prove to be hugely beneficial to developer teams in enterprises.

“For one, it forces the developer to think through the requirements clearly, leading to better outputs from the AI,” said Igor Beninca, data science manager at AI consultancy services firm Indicium.

In addition, Beninca said that the defined spec is always present inside Kiro, providing a durable context for the AI.

“If developer teams need to make a change days later, they modify the spec, and the AI understands the goal within the original context. This is far superior to trying to re-explain everything in a chat window,” Beninca said.

Typically, in other IDEs, developer teams would need to input long-form chat prompts or feed documents into a context window to re-explain the application requirement.

Unique in a crowded space, but few takers for now

Experts and analysts also feel that Kiro, which is entering the fast-evolving field of AI-native vibe-IDEs, alongside tools like Windsurf, Cursor, and Claude Code, is unique.

“While rivals, such as Windsurf and Cursor, are exploring similar ideas with project-wide context and multi-agent collaboration, respectively, Kiro’s formal embrace of specs and hooks appears uniquely mature right now,” Ashmore said.

Ashmore cautioned that while Kiro may appeal to AI-first teams and tool-builders seeking a customizable agentic IDE, it’s unlikely to draw mainstream VS Code users in large numbers, as most developers still rely on traditional environments like VS Code or JetBrains tools.

Kiro’s introduction also comes at a time when the agentic IDE space is seeing a surge in activity: coinciding with developments, such as Google and Cognition dividing up parts of Windsurf and Cursor losing favor with its subscriber base due to a pricing change.

While there’s no exact market share data for vibe-coding or agentic IDEs, recent indicators, such as a survey conducted by The Pragmatic Engineer this year, suggest a clear pecking order, according to Ashmore.

Cursor leads the pack with strong adoption and an estimated $100 million in annual recurring revenue, followed by Windsurf at around $50 million and a growing enterprise base, Ashmore said, adding that Devin, despite its media attention, trails significantly in usage—reportedly one-tenth as popular as Windsurf and just 1% as widely used as Cursor.

Other players, such as Claude Code, Replit’s AI IDE, and Cline, are gaining niche traction but have yet to release concrete metrics, he further said.

Pricing and availability

Just like most AWS services, Kiro will be available at a tiered pricing model and is currently in public preview.

While free during its public preview period with a cap of 50 agentic interactions per month, AWS will later offer paid plans starting at $19 per month for 1,000 interactions and $39 per month for heavier users requiring up to 3,000 interactions.

Rivals like Cursor, Claude Code, and GitHub Copilot offer plans at $20 per month, around $17–$20 per month, and $10 per month, respectively.

GitHub also offers the Pro+ plan at $39.