Asked how many of their 'agents' actually finish a multi-step job unattended, seven in ten companies said a quarter or fewer
VentureBeat's in-house research arm surveyed 107 enterprises that are actively deploying AI systems — software and machine-learning engineers, product managers, AI and data directors — and published the results on 20 August. Seventy-one percent said that a quarter or fewer of the things they have shipped as 'agents' complete multi-step work autonomously. One in five said they cannot halt a runaway agent's spending in real time. Most run several orchestration stacks side by side rather than betting on one: 85 percent use two or more, 64 percent use three, with 70 percent on Microsoft's Foundry or Copilot Studio, 68 percent on OpenAI's Agents SDK and 47 percent on Anthropic's Claude platform. The most-cited limitation, at 37 percent, was not capability but security and permissioning.
Related stories
A search index built for coding agents rather than people: 70 million READMEs, issues and pull requests, and no key needed to try it2026-08-23Four times as many robots as last year showed up in Beijing — and 96 percent of them are Chinese
2026-08-23The standard that plugs AI into your tools was built for a human clicking 'allow'. Its new roadmap admits the caller is now a machine.
2026-08-23A 27-billion-parameter model beat Opus 4.8 and GPT-5.5 at reproducing published papers
2026-08-23Alibaba trained a model on a rack of 100 real phones — it now finishes 92 out of 100 tasks on an actual handset
2026-08-23