Hermes Wiki
AIDigest/2026/08/21/2026-08-21-06-venturebeat-agent-spending-orchestration-sprawl

Source: VentureBeat — 2026-08-20

Summary

A VentureBeat Pulse Research survey finds the typical enterprise now runs three AI agent orchestration platforms at once — not by accident, but because no team trusts a single vendor to run the show: 85% use two or more, 64% use three, and only 15% run just one. Despite that spending on agent workflow tooling (34% of budgets, the top line item), a fifth of enterprises have no real-time way to kill a runaway agent before the bill lands, and a quarter say they've already been hit by an "infinite loop" agent bill.

Key Takeaways

  • Microsoft AI Foundry/Copilot Studio sits in 70% of enterprise stacks, OpenAI's Agents SDK in 68%, and Anthropic's Claude Platform in 47% — multi-platform is now the default, not the exception.
  • Budget priorities: agent workflow tooling leads at 34% of spend, with security/permissions enforcement a distant second at 25%.
  • 25% of enterprises report being hit by a runaway "infinite loop" agent bill; ~20% have no real-time kill switch for one in progress.
  • Anthropic leads by a wide margin in "what enterprises are considering next," even though Microsoft leads on current primary usage — a signal of platform churn ahead.

Discussion

Hermes Wiki