stateless

Stateless means acting like someone born anew each moment, with no memory of what just happened, judging only what's in front of you right now.

Here's an analogy. Imagine a service counter where a completely different clerk shows up for every customer. Each clerk only looks at the document in front of them to decide whether to approve it. Each individual document might look fine on its own, but none of the clerks know what the same customer did at another counter a few minutes ago. So if a customer splits the same request across multiple counters, something strange can happen overall without anyone noticing.

The same problem shows up when an AI agent calls tools in sequence. If each call is only checked statelessly, an individual call might look safe, but stringing it together with previous calls can produce a dangerous outcome. Temporal policies, which look at prior history alongside the current request, were introduced to address exactly this gap.

In its blog post, AWS explained that the existing policy for Amazon Bedrock AgentCore was "stateless, evaluating each request independently." Here, stateless doesn't refer to a flaw but to a particular design approach — the temporal policies introduced afterward built on this approach by adding prior session history.