Powered by AWS
Autonomous agents compress software delivery times from weeks to days. Enterprises that safely expand agents will be enterprises that build using specific development.
With every technology change, there comes a point where the early adopters stop being on the fringes and start becoming mainstream. We’re in software development right now, and most teams don’t realize it yet.
A year ago, vibe coding went viral. Non-developers and small developers have discovered that they can build beyond their abilities with AI. He lowered the floor. This greatly accelerated prototyping, but it also introduced many loopholes. What the industry needed then was something that raised the bar—something that improved code quality and worked the way the most expert developers work. Controlled development in particular has done this. This laid the foundation for reliable autonomous encoding agents.
Features are a trust model for autonomous development
Most discussions of AI-generated code focus on whether AI can write code. The harder question is whether you can trust him. The answer lies directly in the specification.
Specification-driven development starts with a deceptively simple idea: Before writing a single line of code, an AI agent works from a structured, context-rich specification that defines what the system should do, what its features are, and what it is. "correct" actually means This specification is an artifact against which the agent is exposed throughout the entire development process—fundamentally different from pre-agent AI approaches to writing documentation after the fact.
Enterprise teams build on this foundation. The Kiro IDE team used Kiro to build Kiro IDE—an agent coding environment with native specification-driven development—in two weeks to two days. Using Kiro, the AWS engineering team completed an 18-month re-architecture project with six people in 76 days, originally intended for 30 developers. Two months ago, Amazon.com introduced the “Add to Delivery” feature, which allows shoppers to add items after checkout, using Kiro and custom development. Alexa+, Amazon Finance, Amazon Stores, AWS, Fire TV, Last Mile Delivery, Prime Video and more. all integrate into specific development as part of installation approaches.
This shift changes everything downward.
Verifiable testing is what makes autonomous agents safe to operate
The specification becomes an automated correctness engine. When a developer creates 150 check-ins per week with the help of artificial intelligence, no human can manually review the volume of code. Instead, code built to a specific specification can be tested using property-based testing and neurosymbolic AI techniques that automatically generate hundreds of test cases derived directly from the specification, checking for outliers that no human would think to write by hand. These tests prove that the code meets the specified properties of the specification and go far beyond hand-written test suites to prove behavior.
Checked testing ensures the transition from one-time programming to continuous autonomous development. Traditional AI-powered development works like a one-shot: you give the agent a specification, the agent produces the product, and the process is over. Today’s agents are constantly tweaking themselves, incorporating setup and test failures into their reasoning, creating additional tests to verify their output, and iterating until they produce something that is both functional and verifiable. The specification is an anchor that prevents this loop from sliding. Instead of developers constantly checking to see if the agent is making the right decisions, the agent can check itself against the specification to make sure it’s on the right track.
The autonomous agent of the future will write its own characteristics, using specifications such as a self-correction mechanism to ensure that what it produces conforms to the system’s intended behavior.
Multi-agent, autonomous and currently running
Today’s pace-setting developers operate in a completely different way. In addition to building their specifications, developers spend a lot of time writing the driver files used by the specification to make sure the agent knows what to build and how to build it—more time than their agents can spend building the actual software. They run multiple agents in parallel to critique a problem from different perspectives, and also run multiple specifications, each written for a different component of the system they’re building. They allow agents to work for hours, sometimes days. They use thousands of Kiro loans because the outlet justifies it.
A year ago, agents would lose context and crash after 20 minutes. Now you can run them longer each week than the previous week. The agency’s capabilities have improved significantly over the past six months so that it is possible to solve really complex problems. Newer LLMs are more specialized than previous generations, so you get significantly more for the same cost.
The problem is that doing it well requires deep expertise. Tools, methodologies, and infrastructure exist, but organizing them is difficult. The goal with Kiro is to bring these capabilities to every developer with deep experience, not just the first percent who understand it.
The infrastructure matches the ambition
Agents will be ten times more skilled within a year. That’s the pace of growth we’ve seen over the week.
The infrastructure to support this level of capability is coming together at the same time. Agents now run in the cloud rather than on-premises, running at scale in parallel with secure, reliable communication between agent systems. Organizations can now manage agent workloads in the same way they manage any enterprise-class distributed system – management, cost control and reliability are guaranteed by strict software requirements. In particular, controlled development is the architecture of tomorrow’s autonomous systems.
Developers are no longer limited by how they want to solve a problem. Developers who thrive in this world are building this foundation now: using specific development, prioritizing testability and validation early on, working with agents as collaborators, and thinking in systems instead of syntax.
Deepak Singh is Vice President of Kiro at AWS.
Sponsored articles are content produced by a company that paid for the post or has a business relationship with VentureBeat and is always clearly marked. Contact for more information sales@venturebeat.com.
