Amazon Online Services introduced on Wednesday Kiro’s powerssystem that enables software developers to supply their AI coding assistants with immediate, specialized knowledge of specific tools and workflows – eliminating what the company calls a fundamental bottleneck in how AI agents operate today.
AWS made the announcement during its annual event re:Invent a conference in Las Vegas. This feature is a departure from the way most AI coding tools currently work. Typically, these tools load all possible possibilities into memory at once – a process that consumes computational resources and can overwhelm the AI with irrelevant information. Kiro’s powers take the opposite approach, activating specialized knowledge only when the programmer actually needs it.
“Our goal is to provide the agent with specialized context so they can get to the right result faster – and in a way that also reduces costs,” said Deepak Singh, vp of agent and developer experiences at Amazon, in an exclusive interview with VentureBeat.
The launch includes partnerships with nine technology firms: Datadog, Dynatrace, Figma, Neon, Netlife, Postman, Stripe, Supabaseand your personal AWS services. Developers can even create their very own powers and share them with the community.
Why AI coding assistants choke when developers plug in too many tools
To understand why Kira permissions matter, helps understand the growing tension in the market for artificial intelligence development tools.
Modern AI coding assistants rely on something called Model context protocolor MCP to connect with external tools and services. When a developer wants his AI assistant to work with Stripe for payments, Figma for design, and Supabase for databases, he connects MCP servers for each service.
The problem: Each call loads dozens of tool definitions into the AI’s working memory before it writes a single line of code. According to AWS documentation, connecting just five MCP servers can eat greater than 50,000 tokens – or about 40 percent of the AI model’s context window – before a developer writes the first request.
Developers are talking more and more loudly about this problem. Many people complain that they don’t need to burn through token allocations simply to have an AI agent work out which tools are relevant for a specific task. They wish to jump into the workflow immediately, not watch an overloaded agent attempt to sort out irrelevant context.
This phenomenon, which some in the industry call “context rot,” results in slower responses, lower quality results and much higher costs – because AI services typically charge nominal fees.
Inside technology that loads AI expertise on demand
Kiro Powers solves this problem by wrapping three components into a single, dynamically loaded package.
The first component is a control file called POWER.md, which acts as a deployment guide for the AI agent. It tells the agent what tools are available and, most significantly, when to make use of them. The second element is the MCP server configuration itself – the actual connection to external services. The third accommodates optional hooks and automation that trigger specific actions.
When a developer mentions “payment” or “checkout” in a conversation with Kiro, the system routinely prompts the power of Stripe, loading tools and best practices in context. When the developer goes to work with the databases, Supabase prompts and Stripe deactivates. Base context usage when no powers are energetic approaches zero.
“You click a button and it loads automatically,” Singh said. “Once the power is created, developers simply select “open in Kiro” and the program launches the IDE with every thing able to use.”
How AWS brings elite programming techniques to the masses
Singh described Kiro’s rule as the democratization of advanced development practices. Before this capability, only the most sophisticated developers knew easy methods to properly configure their AI agents in a specialized context—writing custom control files, creating precise tooltips, and manually managing which tools were energetic at any given time.
“We found that our developers were adding new capabilities to make their agents more specialized,” Singh said. “They wanted to give the agent special powers to solve a specific problem. For example, they needed a front-end developer and wanted the agent to become an expert in backend as a service.”
This commentary led to a key insight: If Supabase or Stripe could build an optimal context configuration once, every developer using these services may gain advantage.
“Kiro’s powers formalize this – things that humans have done, only the most advanced ones – and allow anyone to acquire these kinds of skills,” Singh said.
Why dynamic loading outperforms fine-tuning in most AI coding use cases
The announcement also makes Kiro’s powers a cheaper alternative to tuning, which is the process of training an artificial intelligence model on specialized data to enhance its performance in specific domains.
“It’s much cheaper,” Singh said when asked about power versus tuning. “Tuning is very expensive and most pioneering models cannot be tuned.”
This is an necessary point. The most effective AI models since Anthropic, OpenAIAND Google are typically “closed source”, meaning developers cannot modify the underlying training. They can only influence the behavior of models through the hints and context they supply.
“Most people are already using advanced models like Sonnet 4.5 and Opus 4.5,” Singh said. “We need to point these models in the right direction.”
The dynamic charging mechanism also reduces running costs. Because permissions only activate when needed, developers don’t pay to make use of tokens in tools they don’t seem to be currently using.
Where Kiro’s power suits into Amazon’s larger bet on autonomous AI agents
Kiro Powers comes as part of AWS’ broader push toward what the company calls “agentic AI” – artificial intelligence systems that may operate autonomously for prolonged periods of time.
Previously at re:Invent, AWS announced three “border agents” designed to run for hours or days without human intervention: the autonomous Kiro software development agent, the AWS security agent, and the AWS DevOps agent. They represent a different approach than Kiro’s powers – solving large, ambiguous problems reasonably than providing specific expertise for specific tasks.
Both approaches complement each other. Border agents handle complex, multi-day projects that require autonomous decision-making across multiple code bases. In turn, Kiro’s capabilities give developers precise, efficient tools for on a regular basis programming tasks where speed and token performance are paramount.
The company is betting that developers need each ends of this spectrum to be productive.
What has Kiro’s power reveals the future of AI-powered software development
The launch reflects the maturing market for artificial intelligence development tools. GitHub Copilot, which Microsoft launched in 2021, introduced hundreds of thousands of developers to AI-powered coding. Since then, there has been an increase in tools – including Cursor, ClineAND Claude Koda — competed for developers’ attention.
However, as these tools became more powerful, in addition they became more complex. The Model context protocolwhich Anthropic opened sourced last yr, created a standard for connecting AI agents to external services. This solved one problem while creating one other: context overload, which Kiro now solves.
AWS positions itself as a company that understands manufacturing software development at scale. Singh emphasized that Amazon’s experience running AWS for 20 years, combined with its own massive internal software engineering organization, gives it unique insight into how developers actually work.
“It’s not something you can just use in a prototype or a toy app,” Singh said of AWS’s AI development tools. “If you want to build production applications, there is a lot of expertise that we bring as AWS that applies here.”
The road ahead is to the power of Kiro and cross-platform compatibility
AWS has indicated that Kiro’s powers currently only work inside Kiro IDEbut the company is working on compatibility with other AI programming tools, including command-line interfaces, Cursor, ClineAND Claude Koda. The company’s documentation describes a future in which developers can “build power once and use it anywhere” – although for now this vision stays aspirational.
For technology partners launching permissions today, the appeal is easy: Instead of maintaining separate integration documentation for every AI tool available on the market, they will create a single solution that works wherever Kiro does. As more AI coding assistants enter the market, this sort of performance becomes more and more worthwhile.
Kiro’s powers are there available now for developers using Kiro IDE version 0.7 or later, at no additional cost beyond the standard Kiro subscription.
The basic bet is a familiar one in the history of computing: the winners in AI-powered development is not going to be tools that attempt to do every thing at once, but people who are smart enough to know what to forget.