AWS unveils million-token context, 14-day agents, Strands Robots

AWS's August 2026 announcements reflect a shift from raw model performance to the full system around AI, including context, action, geography, and governance. For example, OpenAI's GPT-5.6 family now handles million-token contexts—equivalent to entire codebases or regulatory documents—and can retrieve fresh web content in a single API call. Cross-Region inference adds global routing capacity, and lower per-token prices for the GPT-5.6 family on Bedrock signal efforts to manage cost as usage scales.

The expansion of AgentCore to run sessions up to 14 days on dedicated compute suggests agents are taking on longer-horizon research, coding, and monitoring tasks. Temporal policies and rate limiting provide the necessary guardrails for such autonomy. Strands Robots bridges AI agents to physical hardware, a step toward broader deployment, but its limited research preview indicates early stages.

The stakes for AWS appear to be maintaining its position as the infrastructure layer for both digital and physical AI systems. The challenge will likely be balancing autonomy and control as agents operate across longer periods and broader boundaries. Adoption hinges on whether enterprises trust the governance tools—cost allocation, anomaly detection, and memory isolation—enough to delegate consequential work.