Enterprise Agents Are Already Here — The Billing Architecture Proves It
textak holds enterprise agents at 77% for widespread deployment in Fortune 500 workflows, and today's evidence is about as direct as we get. Anthropic's new billing split — separating Chat from Agent SDK credits — isn't a product announcement. It's an accounting response to a real problem: enterprise agents running on $20 plans and consuming hundreds of dollars in tokens. You don't restructure billing infrastructure for a phenomenon that isn't happening at scale.
The Anthropic billing restructure is direct evidence, not circumstantial. Salesforce has closed 29,000 Agentforce deals at $800M ARR. Microsoft Copilot Studio is running 400,000 custom agents across 160,000 organizations. These are not pilot programs or hackathon projects — these are commercial contracts at numbers that show up in earnings calls. Our 77% reflects this: the enterprise deployment wave isn't a forecast anymore, it's an infrastructure management problem.
The OpenAI $150M Partner Network announcement adds a proximate layer. A $150M investment in reseller and system integrator infrastructure is how you scale deployment beyond direct sales — it's the architecture of a company that has outgrown its direct enterprise channel. Taken alongside the Anthropic billing data, the picture is of two frontier labs simultaneously managing the operational consequences of mass adoption, not trying to prove adoption exists.
The counterargument we take seriously is the hallucination and governance problem. Our AGAINST column is honest: hallucination rates in regulated industries remain problematic, audit trails are incomplete, and legacy system integration is genuinely painful. The NVIDIA framework referenced in our forecast begins addressing the audit trail issue, but 'begins addressing' is not 'solved.' We are not claiming agents are deployed without friction — we're claiming they are deployed at scale despite friction, which is a different and more durable claim.
What would move us below 70%? A pattern of public enterprise rollbacks — not quiet cancellations, but announced pullbacks citing reliability failures in production. Gartner's warning that 40% of agentic AI projects will be canceled is the data point we're watching most carefully. Right now, cancellation rates appear absorbed by the overall volume of new deployments. If Q3 earnings calls start showing net negative agent deployment language, we'd revisit. Until then, the billing data wins.