Enterprise Agents Are Already Here. The Billing Infrastructure Proves It.
textak holds 77% on autonomous agents reaching widespread enterprise deployment, and today's evidence is about as direct as we get in this business. Anthropic didn't restructure its billing because agents are a promising pilot program — it restructured because agents running on $20 consumer plans were consuming hundreds of dollars in tokens at scale. When a company rewrites its pricing architecture to combat 'subscription arbitrage,' that's not a signal that production deployment is coming. That's a signal it already arrived.
Let's be precise about what the Anthropic billing split actually proves. The new structure — separating Chat usage from Agent SDK credit pools effective June 15 — is a direct operational response to enterprise-scale agent consumption overwhelming consumer-tier plans. This is direct evidence of production deployment, not proximate evidence that conditions exist for it. You don't architect billing pools around a theoretical future problem. Pair this with Salesforce closing 29,000 Agentforce deals at $800M ARR and Microsoft Copilot Studio running 400,000+ custom agents across 160,000 organizations, and the deployment picture is unambiguous at scale.
Our 77% reflects the compounding of three things: major cloud providers shipping agent frameworks into enterprise sales channels, efficiency claims (40%+ in pilots) that are generating real procurement decisions, and now billing infrastructure that confirms consumption at a level worth protecting. The move from 76% to 77% was incremental because this confirmatory evidence, while strong, doesn't resolve the forecast's hardest edge — 'widely deployed' still begs the question of regulated industries. Healthcare and financial services remain structurally different from the tech-adjacent enterprises leading adoption.
Here's the counterargument we take seriously: 400,000 custom agents across 160,000 organizations sounds large, but averages out to 2.5 agents per organization. Many of these are almost certainly task-narrow automations — glorified scheduled scripts with an LLM attached — rather than the autonomous multi-step workflow agents the forecast envisions. Salesforce's 29,000 Agentforce deals are similarly real, but Salesforce is not a neutral reporter of its own adoption. The $800M ARR number is verifiable; whether those deployments constitute 'autonomous agents in workflows' versus 'AI-assisted CRM features' is harder to audit independently.
What would push us toward 85%+: a major regulated-industry deployment announcement — a top-10 bank or health system publicly describing agents handling end-to-end workflows without human review gates, with specific process names and headcount implications. What drops us below 65%: evidence that the majority of 'agent deployments' being counted in these metrics are single-tool, single-step automations that don't meet any reasonable definition of autonomous workflow execution. The Anthropic billing restructuring is the most honest signal we've seen — companies consuming enough agent tokens to require infrastructure redesign aren't running experiments.