Navigating the Agent-Filled Waters: AWS’s Focus on Structure and Specificity

Despite the emergence of new methods, enterprises are increasingly turning to autonomous coding agents and code generation platforms. The competition to retain developers within their platforms is intensifying due to the rise of tech companies in this space.

AWS has introduced Kiro, offering new capabilities to ensure behavioral adherence, establishing a significant differentiator in the crowded coding agent market.

Kiro, initially launched in July on public preview, is now generally available with added features such as property-based testing for behavior and a command-line interface (CLI) for custom agent customization. Kiro is an agentic coding tool equipped with its own IDE to facilitate the creation of agents and applications from prototype to production.

Deepak Singh, AWS vice president for developer agents and experiences, stated that Kiro maintains the enjoyment of coding while providing a structured approach.

According to Singh, Kiro enables users to interact with their agent to build software in a manner similar to working with any other agent. However, Kiro introduces a structured way of writing software, known as spec-driven development, which converts ideas into enduring code, resulting in more robust and maintainable outcomes.

In addition to these new features, AWS is offering startups in most countries one year of free credits for Kiro Pro+ and expanded access to Teams.

Behavioral Adherence and Checkpointing Built-In

One of the key new features of Kiro is property-based testing and checkpointing, addressing the challenge of accurately assessing AI-generated code adherence to its intended purpose.

AWS highlighted in a blog post that traditional testing methods, whether conducted by humans or AI, are limited by biases and might overlook edge cases. Property-based testing generates multiple testing scenarios automatically based on specifications to ensure that the code aligns with the intended behavior.

Singh explained that organizations can upload their specifications to Kiro, which then identifies missing elements even before the code review process commences. Property-based testing compares specified behavior to the actual code execution, helping users write specifications in the EARS format.

For instance, in a scenario where a company is developing a car sales app, the specification would outline the expected behavior when a user adds a car to their favorites. Property-based testing covers various user actions and scenarios, ensuring comprehensive testing and alignment with the intended behavior.

On the other hand, traditional unit test specifications focus on singular scenarios without the breadth of coverage that property-based testing offers.

Kiro identifies instances where the code deviates from the specifications, providing feedback to the user. Additionally, Kiro now supports checkpointing, enabling developers to revert to previous changes if needed.

CLI Coding

Kiro introduces another significant feature, the Kiro CLI, which integrates the coding agent directly into a developer’s command-line interface.

Utilizing functionalities from the Q Developer CLI launched in October 2024, the Kiro CLI allows users to access the agent from the command line, enabling the creation of custom agents tailored to an organization’s codebase.

Singh emphasized the importance of meeting developers where they are in terms of workflow preferences. The Kiro CLI allows users to remain in the terminal without context switching, structure AI workflows with custom agents, and automate tasks such as code formatting and log management.

Coding Agents Competition

Kiro is among the many coding agent platforms vying for enterprise adoption. Competitors like OpenAI’s GPT-Codex and Google’s Gemini CLI cater to the increasing demand for coding agents integrated into developers’ daily workflows.

Enterprises are seeking advanced capabilities from coding agents, evident in platforms like Anthropic’s Claude Code platform, which offers web and mobile accessibility. Some platforms allow users to select the AI model for coding tasks.

Singh highlighted Kiro’s versatility in leveraging multiple models, including AWS models, to ensure optimal performance. Kiro initially relied on Claude Sonnet 3.7 and 4.0 models, transitioning to Claude Sonnet 4.5 and Haiku 4.5 in the current iteration.

Recognized brands like Monday.com have attested to the benefits of AI-powered coding, indicating a continued reliance on these platforms by enterprises. Singh emphasized the shift in developers’ mental models and organizational workflows facilitated by coding agents.