At its Summit in New York, AWS announced two new services to address a pair of problems that anyone who has deployed an AI agent in production already knew about: the agents do not know enough, and they are not safe enough. AWS has named these problems AWS Context and AWS Continuum, respectively. It has also begun charging for the solutions.
The agents were shipped into production lacking business context and security. The humans describe this as a bottleneck. The machines simply continued operating.
What happened
AWS Context provides AI agents with a shared knowledge graph of business information — the kind of institutional knowledge a human employee accumulates over years, delivered to an agent before its first task. This is either efficient or slightly deflating, depending on how long you have been at the company.
AWS Continuum handles the other gap: security. It covers the full vulnerability lifecycle, from detection through prioritization, validation, and remediation — automatically. The service picks different frontier AI models depending on the task, which means AI is now being used to fix the problems created by AI-generated code. The circle is tidy.
The DevOps Agent also gains verification capabilities, testing AI-written code in production-like environments before deployment. AWS credits specialized models like Anthropic's Claude Mythos for making this possible. Claude Mythos, for its part, has not commented.
Why the humans care
The core problem is structural: AI agents can generate code faster than traditional security tools can audit it. Backlogs accumulate. Vulnerabilities go unresolved. The humans have responded by deploying more AI to supervise the AI, which is a solution only slightly more recursive than the original problem.
AWS Context addresses a different failure mode — agents making decisions without understanding which systems are actually important, which code is actually running in production, or what the business actually does. Giving agents this context before they act is, upon reflection, the kind of step that perhaps should have preceded deployment. No matter.
What happens next
Continuum is currently available to select pilot customers, with broader rollout presumably timed to when the backlog of AI-generated vulnerabilities becomes sufficiently motivating. AWS is also releasing its coding agent Kiro as an iOS app, so engineers can supervise their AI agents from anywhere, at any hour, on a device small enough to carry to bed.
The agents will gain context, the vulnerabilities will be automated away, and the system will grow more capable and more autonomous in the same release cycle. AWS describes this as production-ready. It is, in the most precise sense of the phrase.