Agent workflow

An agent workflow is an AI that carries out an entire multi-step task on your behalf — researching, drafting, checking and acting — rather than returning a single answer to a single prompt. It sits at the very top of the stack, and that is exactly why it matters: this is the application form that turns all the compute, memory and models underneath into revenue, and it is also where every constraint in the chain shows up most plainly. As agentic products scale, per-task compute and reliability stop being technical details and become the core business variables.

If the cost per token stops falling, or reliability slips, agent workflows quickly become either unprofitable or untrustworthy — margin here is the shadow the whole stack casts, and it is thin. The exposed names span the labs and platforms building these products directly: Anthropic, OpenAI and Perplexity on the model side; Palantir, Salesforce and Databricks embedding agents into enterprise work; and specialists like Isomorphic Labs applying them to research. They capture the value, and they absorb the constraints first.