Enterprises Struggle to Halt Runaway AI Agent Spending in Real Time

Runaway AI Agent Spending: Why One in Five Enterprises Can’t Contain Costs

According to the latest VB Pulse analysis, runaway AI agent spending remains a blind spot for approximately 20% of large firms, which cannot cut off a rogue agent in real time. Enterprises are abandoning the notion of a single orchestration stack; the median organization now juggles three platforms, a strategy driven by doubts over vendor security and permission controls. Reuters notes that multi‑vendor deployments have risen sharply, with ↑ 70% of respondents citing Microsoft AI Foundry or Copilot Studio and ↓ 15% expressing frustration over implementation ease.

Hybrid Control Planes Take Center Stage

More than half of surveyed firms expect a hybrid control layer by 2026, while 14% plan to rely on provider‑managed services and 13% on custom in‑house solutions. The data show a rapid churn: 15% will switch platforms within three months, another 24% in the next half‑year, and 28% within a year, with Anthropic’s Claude SDK topping the shortlist.

“We’re building for an ecosystem where models, tools and orchestration layers interoperate, not for a single winner,” a senior AI director told Bloomberg.

Spending priorities have shifted toward monitoring (31% of budgets) and permission enforcement (30%). Yet only 23% rate overall satisfaction at 4.2/5, while ease of rollout lags at 3.9/5. The gap signals that many enterprises are still in the early stages of true multi‑step agent deployment, with just 2% claiming fully autonomous pipelines.

Words by: Elara Vance

Night-Shift Breaking News Lead