Early pilots show 75% of inbound issues resolved without human involvement; deployments require OpenAI Forward Deployed Engineers — there is no self-serve option.
Why it matters: Presence validates the managed-agent-for-CX thesis — early enterprise adopters will need an implementation partner to configure policies, evaluation sets, and escalation flows before it goes into production.
The hf_fs tool provides a single interface to repos, storage, docs, and papers in ~1,000 tokens; Sandboxes attach secure execution environments to Hub buckets for dataset analysis, training, and Space creation.
Why it matters: As open-source models become part of production agentic stacks, better MCP tooling means tighter integrations and lower token overhead on Hub queries — useful for teams evaluating alternatives to frontier APIs.
A 50-turn session was consuming 2,500x more compute than a 10-turn session due to message normalization costs growing quadratically; anyone running extended agentic loops — CI pipelines, deep refactors, research sessions — should update immediately.
Why it matters: Developer teams running Claude Code for production workflows should update now — the fix directly reduces token burn on any session that runs long.