Agents

Agentic systems and the infrastructure demands they create.

4 articles

Cost control1

Reliability3

About Agents

What is infrastructural about agents?

The cost and reliability consequences. An agentic workload turns one request into many model calls with retries and tool use in between, which changes the token bill, the latency budget and the failure modes of everything downstream. Coverage is about running these systems rather than about designing the prompts inside them.

What does production readiness mean for an agent?

Being able to answer what it costs per completed task, what it does when a step fails, and how its behaviour is observed once it is live. Reports here work through those questions against published guidance and documented deployments, and are explicit about which parts remain unmeasured across the field.