Foundation Models Enable Workflow Platforms

This points to the core split in AI, where model companies sell general intelligence and app companies turn that intelligence into job specific software. OpenAI can supply the base model, APIs, agent tools, and enterprise runtime, but the real product work is teaching the system a banker workflow, a legal review flow, or a support queue, then wrapping it in the right UI, approvals, and data connections. That is why the platform model fits, it lets the model provider collect usage across many apps without owning every workflow.

• OpenAI increasingly looks like infrastructure. Its developer platform explicitly sells APIs and SDKs for others to build on top, and Frontier is positioned as the layer for deploying and managing agents inside enterprises rather than as a finished app for every role.
  2 sacra 3 openai
• The pattern matches AWS. AWS won by giving developers compute, storage, and tooling, then letting thousands of software companies build the vertical products. The analogy here is that foundation models are the new raw capability, and the workflow company captures demand by shaping that capability into a task people already do every day.
  4 openai 5 openai
• The market is already converging this way. Glean, Zapier, Airtable, Writer, and others are colliding around enterprise AI because the value is moving from raw chat to systems that connect company data, call tools, and complete multi step work inside a specific product context.
  1 sacra 6 openai

Going forward, the biggest winners are likely to be the companies that own the workflow layer while staying flexible on models underneath. Model platforms will keep taking a broad tax on usage, but the durable application companies will be the ones that turn generic reasoning into repeatable work inside sales, law, finance, support, and other concrete jobs.