AWS aims to accelerate agentic AI with ‘frontier’ technology

As we head into 2026, AI is preparing to boldly go where no agent has gone before—the final frontier.

Well, at least in part. Frontier agents are a new expansion of AI’s capabilities, pushing agentic AI to new autonomous horizons. This new generation of agents could work on complex tasks for hours, or even days, without needing a human in the loop, AWS claims.

That’s the next step for the technology, said Deepak Singh, AWS VP of developer agents and experiences.

“The agent is always getting more knowledgeable, it’s learning more and more the fine details of how you want to get things done,” Singh told IT Brew. “It can work alongside you or as an extension of you.”

AWS has developed frontier agents that focus on three components—autonomy, scalability, and independence. The agents will focus on software development and each have a specific task:

• The Kiro autonomous agent is focused on coding
• The AWS Security agent is devoted to risk assessment and incident reporting
• The AWS DevOps agent, is all about, well, DevOps

Frontier agents could deploy for hours or days at a time without intervention, Singh said, but that’s only if everything goes right. The agents will stop operations if they need human guidance or have a problem they can’t solve.

“If you’re giving it a task, you have full visibility into what it’s doing,” Singh said.

With frontier agents, AWS joins a growing number of tech companies like IBM and Microsoft trying to take AI agentic technology to the next level. Even Walmart is debuting “super-agents.”

Businesses hoping to jump on the hot new commodity and make a mark is a familiar story, but AI agents are unique in that they can both have longstanding effects on the user experience and add to security concerns. Singh said that users will maintain control over permissions; agents involved in security, for example, have detection capabilities but that doesn’t mean they can take rehabilitative actions.

“If it finds a security issue, it’s not going to arbitrarily go change the code—it’s going to file a ticket and assign an issue which somebody else can pick up to fix before the code can be moved forward,” Singh said, noting that these guardrails are “all built into the agent design.”