Kiro has launched Kiro Web in preview for paid subscribers, with the company stating that the browser-based product can "write code, coordinate changes across repositories, and open pull requests." Access is limited to Pro, Pro+, and Power plans, and Kiro also says Web credits are 50% off until May 29, 2026.
The rollout follows the company’s re:Invent announcement around autonomous software development. Kiro now presents that capability as part of the core experience rather than a separate product, at least in the company’s own description. Support for autonomous workflows in Kiro IDE and Kiro CLI is scheduled to arrive in the coming weeks.
How Kiro Web works
Kiro Web starts with a session in which repositories are selected and a task is described. From there, users can discuss implementation approaches, ask questions about the codebase, iterate on changes, and request a pull request once the work is ready.
The company also states that Kiro brings context from repositories, web search, and prior code reviews into responses. When a task is handed off end to end, an "Autonomous" toggle prompts Kiro to ask clarifying questions, build a plan, and coordinate sub-agents for planning, coding, and verification. The result, according to Kiro, is a pull request rather than a progress update.
Multi-repo workflows and GitHub integration
Kiro Web is designed to work across multiple repositories in one session. The company says that a single task can span a shared library and the services that depend on it, or a backend change and the frontend client that consumes it.
The product also ties into GitHub in two ways. Issues can be assigned to Kiro by adding the kiro label or mentioning /kiro in a comment. Once a pull request is open, reviewers can leave comments as usual, and Kiro states that it can respond to feedback, including with /kiro all to address all reviewer comments or /kiro fix to focus on a specific thread.
Steering files and learned feedback
Kiro Web reads steering files that define coding conventions, architecture patterns, technology preferences, and testing standards. These can be uploaded through Kiro's agent settings or imported from a GitHub repository. The product also picks up steering files stored in the .kiro/steering/ folder in repositories.
Kiro says the same steering setup applies across Kiro Web, Kiro IDE, and Kiro CLI. The company also claims the system learns from feedback during sessions and on pull requests, so comments such as "always use our standard error handling" can shape future tasks across repositories.
Isolated cloud sandbox
Each task runs in an isolated sandbox that Kiro describes as fresh for the session and deleted when the work ends. The agent clones repositories into that environment and only has access to resources explicitly permitted by the user or organization, according to the company.
Kiro also states that sessions are isolated from one another and from the local environment. Sandbox controls for network access, MCP tool configurations, and base image settings are available from the settings page. For organizations using AWS Identity Center, administrators can enable or restrict access for their organization, and Kiro points to an Identity Center guide for details.
Getting started
Kiro’s setup steps are straightforward: sign in at app.kiro.dev, connect a GitHub account in Settings, and select repositories before starting a session. If an organization uses AWS Identity Center, an administrator must enable Kiro Web before access is available.
What comes next
Kiro says specs are coming to Kiro Web, with a workflow that would move from requirements and design into implementation by the autonomous agent. The company also states that autonomous workflows are on the way to Kiro IDE and Kiro CLI in the coming weeks.
Kiro Web remains in preview, and the company is inviting users to try it on a real task from a backlog, a bug, or a test suite that has not yet been written.
Source: Kiro
