Evaluating agents before letting them loose
A demo that goes well is not a test. How to build an evaluation set from real cases, which metrics matter and what threshold somebody must sign before granting an agent autonomy.
Palantir Foundry, AIP and Data Strategy Thought Leadership
Latest articles on Palantir and data strategy
Agent FinOps: the cost per transaction
The monthly bill hides what matters: how much each task an agent resolves actually costs, and whether that figure stays below the value it creates. Four levers to control it.
Selecting Your Software Vendor in 30 Days
Software selection should take weeks, not months. Phil Turton, founder of Viewpoint Analysis, explains why 30 days is a realistic target and how the month runs, from a standing start to a preferred vendor.
The ontology as the agent's context
Agents do not fail for lack of model, they fail for lack of context. What an ontology adds over a table schema, and how to build one from a concrete use case.
AgentOps: observing the reasoning, not just the result
Storing an agent's answer will not help you debug it. What to instrument at every reasoning step and tool call, and which signals warn you before something breaks.
An agent is an identity: permissions, scope and owner
Once an agent stops answering and starts acting, the question is no longer what it knows but what permissions it holds. How to treat every agent as an identity with its own scope and a named human owner.