DataRobot vs NVIDIA NeMo

The feed is running a sustained essay series on agent governance, published on a fixed cadence: borrowed credentials give an agent every permission its author holds, credentials should never reach the model, agent identity must be a lifecycle rather than a one-time setting, delegation chains create confused-deputy exposure, and governing five agents differs structurally from governing five hundred. Interleaved with the series are two product-adjacent items — OpenCode, a coding agent that lets teams choose the model behind it, and an executive argument that existing predictive AI infrastructure is the shortest path to agentic value.

The series is building a purchasing argument from first principles: if an agent can act rather than merely answer, then identity, delegation, and scoped authority become the controls that matter, and those are platform concerns rather than model concerns. That framing points squarely at DataRobot's installed base — customers with production models, pipelines, and governance already in place are told they are further along than they think. OpenCode fits the same thesis from the developer side, treating model choice as a policy decision rather than a vendor lock.

Expect the governance series to resolve into a named product surface for agent identity and delegation, since the essays keep describing requirements — stable runtime principals, credential isolation, scoped authority across trust domains — in terms specific enough to be a spec.