z = 3.93 sources4 articles2026-09-01 (±1 day)/en/ai/story/nvidia-2026-09-01
✎ Story body
NVIDIA slotted a science inference component into Claude's agent environment this week, alongside separate moves over where and how much agents get to run.
What happened
NVIDIA published a walkthrough for running BioNeMo NIM microservices inside Claude Science for protein structure prediction. The same cluster carried JetBrains' locally-running Junie agent, a revision to Claude Code's weekly limits, and enterprise rollout coverage — one vendor post against three trade reports.
Why it matters
A GPU vendor shipping a component into someone else's agent platform, rather than a finished model, hints that the contest is shifting from models to the runtime around them. The subject matter is drug discovery, but the stake sits on the developer side. How widely this gets adopted is not yet established.
What to watch
Whether NIM shows up in workflows beyond the official guide. Where local execution and quota redesign settle is the other read on whether always-on agents become the default assumption.