Introduction
As software systems grow in size and complexity, development teams increasingly need tools that support planning and coordination across entire codebases rather than isolated code suggestions. Delivery challenges today are less about writing individual functions and more about keeping intent, implementation, and change aligned over time.
AWS Kiro reflects this shift in tooling. Designed as an agentic IDE, it combines structured specifications with automated execution to support more predictable and governed software delivery.
What AWS Kiro is
AWS Kiro is an agentic integrated development environment built by Amazon Web Services. Based on Code-OSS, it retains the familiar VS Code-style experience while introducing a spec-driven workflow that treats requirements, design, and tasks as first-class artefacts.
Rather than relying solely on conversational prompts, Kiro uses structured markdown files to guide automated actions across the delivery lifecycle. It runs on Amazon Bedrock and currently uses Claude Sonnet models, abstracting model selection and configuration away from development teams.
Kiro is positioned not as a replacement for existing engineering practices, but as a way to formalise them and embed them directly into day-to-day development work.
From AI assistance to AI agency
Early AI coding tools focused on inline suggestions and function-level completion. These approaches work well for isolated changes but struggle with complex systems that span multiple services, repositories, or compliance constraints.
Agent-based tools address this gap. Instead of reacting to individual prompts, they operate against defined goals, plan work in stages, and revisit earlier decisions based on outcomes. This makes them better suited to long-lived codebases and coordinated delivery efforts.
AWS Kiro reflects this shift by prioritising planning and structure before execution.
How AI agents operate in practice
In Kiro, agents work against explicit intent captured in specifications. A request to add a new feature does not immediately result in generated code. Instead, the agent first produces or updates requirements, proposes a design, and breaks the work down into executable tasks.
Once this structure is in place, agents can:
- apply changes across multiple files,
- run and interpret tests,
- update documentation alongside code,
- and iterate based on results.
This behaviour differs from traditional automation. Scripts and pipelines follow predefined steps, while agents adapt their actions based on feedback, within defined guardrails and approval points.
90% reduction in deployment time and 2x increase in operating speed through a large-scale migration to the cloud
Spec-driven development as a stabilising layer
Spec-driven development is central to how Kiro works. Each feature is grounded in three artefacts:
- Requirements capture user stories, acceptance criteria, and non-functional expectations.
- Design documents describe architecture decisions, data models, and integration patterns.
- Tasks break implementation into traceable, reviewable steps.
These artefacts are not passive documentation. They actively guide implementation and remain linked to the code as it evolves. As a result, intent, design decisions, and delivery stay aligned over time.
For teams operating in regulated or high-risk environments, this approach improves traceability, reviewability, and onboarding. New contributors can understand a system by reading specifications rather than reverse-engineering behaviour from source code alone.
Implications for engineering teams
Delivery speed and predictability
By separating intent from execution and automating well-defined work, teams can reduce iteration cycles without losing control. The largest gains tend to come from consistency rather than raw speed.
One of the users of AWS Kiro, CTO at ITS, points to this impact clearly: a modernisation effort estimated at 52 weeks was delivered in three weeks, resulting in a 90% increase in efficiency after introducing Kiro.
Governance and compliance
Structured artefacts make it easier to involve architects, security teams, and compliance stakeholders early. Versioned specifications and traceable links between requirements, tasks, and code changes create a reviewable audit trail, while guardrails that ensure automated actions stay within agreed architectural, security, and regulatory boundaries.
The evolving role of the engineer
With agent-based tools, engineers spend less time on repetitive implementation work and more time on system design, validation, and decision-making. Engineers must actively decide which tasks are suitable for automation and where human judgement is essential, reinforcing the importance of domain expertise and architectural thinking.
How AWS Kiro compares to other AI coding tools
AWS Kiro enters a crowded landscape that includes GitHub Copilot, Amazon Q Developer, Cursor, and similar tools. Most focus on assisting with code authoring inside existing workflows.
Kiro takes a different approach. It is process-first rather than prompt-first, treating the development lifecycle as a coordinated sequence of activities rather than a series of isolated interactions. Specifications are not side effects of development but its primary inputs, and agents operate across files and stages to maintain consistency from planning through to implementation.
This design reflects enterprise realities. Kiro is built with native IAM integration, clear security guardrails, and deep AWS alignment, making it suitable for teams operating in governed environments. At the same time, this structure introduces trade-offs. Kiro is designed for teams rather than individual developers and can feel restrictive for those who prefer full manual control or highly exploratory workflows. Its tight coupling with AWS also makes it a stronger fit for organisations already committed to the AWS ecosystem.
In practice, many teams will use Kiro alongside other tools. Inline assistants remain useful for local changes and experimentation, while Kiro supports structured feature development and cross-cutting work that benefits from shared process and control.
Get recommendations on how AI can be applied within your organisation.
Explore data-based opportunities to gain a competitive advantage.
Where this approach fits and where it does not
Agentic, spec-driven tools are most effective in environments with:
- complex systems and clearly defined problem boundaries,
- team-based development with multiple contributors,
- mature engineering practices and established workflows,
- regulatory or audit requirements that demand traceability,
- and a need for long-term maintainability.
They are less effective for exploratory work where intent is unclear or rapidly changing. In those cases, lightweight tools and manual iteration may still be more appropriate.
Read more about AI on our blog:
Considerations and risks
Agent-based, spec-driven tools change how control is exercised in software delivery. Instead of defining every step, engineers define intent, constraints, and review points, which requires a shift towards a deliberate “trust and verify” model.
Using these tools effectively depends on understanding what can be delegated to agents and what must remain under direct human ownership. Poorly defined intent or weak specifications quickly lead to poor outcomes, regardless of tooling.
There is also an unavoidable loss of strict determinism. Agent-based systems may reach correct results through different paths, making guardrails, approval checkpoints, and traceability essential, particularly in regulated or production environments. At the same time, teams need sufficient system design and domain expertise to review and validate agent-generated changes with confidence.
What this means for AI-enabled modernisation
Agent-based tools such as AWS Kiro fit naturally into broader modernisation initiatives, particularly where organisations are rethinking how they design, build, and operate software at scale. They support clearer interfaces, stronger documentation, and more disciplined delivery, which are often prerequisites for successful cloud and AI adoption.
In this context, AWS Kiro is complementary to AWS Transform. Kiro supports how software is built and maintained, with AI agents embedded directly in the development loop. AWS Transform focuses on modernising existing systems, helping teams lift and shift, refactor, or replatform workloads more safely and efficiently than manual approaches alone. Used together, they address different stages of modernisation, from transforming legacy foundations to sustaining and evolving modern platforms.
From a delivery perspective, this is where experience matters. As an AWS Partner, Future Processing works with organisations modernising legacy systems, building cloud-native platforms, and introducing AI into existing delivery models. This includes defining spec-driven workflows, integrating agent-based tooling into established CI/CD pipelines, and putting governance in place so automation improves quality rather than increasing risk.
In practice, AI-enabled modernisation is not about introducing new tools in isolation. It requires alignment between architecture, delivery processes, cloud platforms, and team capabilities. Agentic IDEs like Kiro can support this shift, but only when combined with clear ownership, strong engineering practices, and an understanding of how AI fits into long-term system evolution.
Summary
AWS Kiro illustrates a broader shift towards agent-based, spec-driven software delivery, where structured intent guides automated execution across the lifecycle. For engineering teams, this can improve consistency, traceability, and delivery confidence, particularly in complex or regulated environments. At the same time, it requires a different operating model, built on clear specifications, governance, and informed human oversight.
I had the pleasure of discussing these topics during a four-part conversation at the Andersen Summit in Las Vegas with Drew Danner, CISSP, PMP, Managing Director and Security Expert at BD Emerson. You can watch it on my LinkedIn:
- The SDLC in the age of AI – watch here
- How security work has fundamentally changed with AI adoption – watch here
We talked about how the introduction of LLMs into software delivery changes both how systems are built and how they must be secured. Topics ranged from securing next-generation applications, including so-called vibe coding, to the reality that legacy systems do not become safer simply because AI is involved.
The discussion also highlighted where AI delivers measurable impact and where its limits remain. Security workflows can be accelerated by 50–60%, and well-designed automation can reduce cost without sacrificing outcomes. At the same time, customised AI agents only add value in clearly defined scenarios, and experienced engineers remain essential for judgement, validation, and accountability.
As one conclusion from that conversation put it:
AI raises the ceiling, but skill determines the outcome.
Keep your business at the forefront of cloud innovation, maintaining cost efficiency, mitigating risks, and ensuring regulatory compliance.
FAQ
What is AWS Kiro and how does it differ from GitHub Copilot or Amazon Q Developer?
AWS Kiro is an agentic IDE built around spec-driven development. While Copilot and Amazon Q Developer focus on inline assistance, Kiro uses structured artefacts to coordinate work across planning, implementation, testing, and documentation.
Is AWS Kiro generally available?
AWS Kiro is currently available in public preview with usage limits. Access and features may evolve as the product matures.
Does AWS Kiro replace developers?
No. Kiro automates well-defined tasks but relies on engineers for design decisions, validation, and accountability. The role of the developer shifts towards steering and review rather than manual execution alone.
How is spec-driven development in Kiro different from traditional documentation?
In Kiro, specifications actively drive implementation and remain linked to the code. They evolve with the system rather than acting as static reference documents.
Which AI models does AWS Kiro use, and how do they differ?
AWS Kiro runs on foundation models provided through Amazon Bedrock. In practice, this includes Anthropic’s Claude family, such as Claude Haiku, Claude Sonnet, and Claude Opus, with different variants offering trade-offs between speed, cost, and reasoning depth.
Lighter models are typically suited to fast, repetitive tasks, while larger models provide stronger reasoning for complex planning, cross-file changes, and specification-driven work. Model selection and orchestration are handled within the AWS ecosystem, so teams do not manage models directly, but benefit from different capabilities while remaining within AWS security, IAM, and governance controls.
