Shift to Task-Based AI Billing

This is a shift from selling answers to selling completed work. Once an agent can click through software, call tools, and finish a task like filing an expense report or updating a CRM record, the spend no longer looks like a seat for software or a salary for junior ops staff, it starts to look like usage based infrastructure for office work. OpenAI is positioning itself to meter that work one task at a time.

• The product mechanics matter here. OpenAI now has agent capabilities that can browse websites, edit spreadsheets, run code, connect to data sources, and use tools through the API. That makes the monetizable unit a finished workflow, not just a chat prompt.
  1 sacra 2 openai 5 openai
• A close analogue is BPO and outsourcing. Invisible already sells outcome based digital labor, like annotations, onboarding, or expense handling, using a mix of software and human agents. AI agents attack the same budget line, but with software gross margins if reliability gets high enough.
  4 sacra
• The clearest wedge is in high value white collar workflows. Harvey shows how this lands in practice, selling legal work reduction on a per seat basis, while OpenAI is training for finance, legal, and consulting tasks directly. That pushes model labs closer to owning domain workflows, not just supplying the underlying model.
  1 sacra 6 sacra

The next phase is enterprises moving from copilots to managed agent fleets. As reliability, governance, and security improve, more spending will move out of labor budgets and fixed SaaS seats into variable AI execution spend, with the biggest winners being the platforms that can own both the model and the workflow layer.