New Enterprise Survey: Businesses That Install OpenAI's Agent Platform Make It Primary — Anthropic Lags Behind

Businesses Are Trying a Lot of Platforms, But Not All of Them Turn Into Commitments

After a week dominated by AI safety debates and calls to slow down development, today brings something different: hard business data on how companies actually choose infrastructure for AI agents they run in day-to-day operations.

The headline finding: According to VentureBeat's VB Pulse survey (August 2026 wave, 169 enterprises with 100+ employees), of 75 companies that have installed OpenAI's Agents SDK or Responses API somewhere in their stack, 52 (69%) named it their primary orchestration platform. Of 45 companies using Anthropic's Claude Platform, only 17 (38%) did the same — the lowest primary-platform rate of any platform with at least 25 installations in the survey.

On overall primary-platform share: OpenAI leads with 33% of enterprises naming it primary, Google's Enterprise Agent Platform follows at 24%, and Anthropic trails at just 11%.

But the picture is more nuanced: Anthropic leads on a different metric — the ratio of forward consideration to current install base. Of 121 enterprises naming platforms they're considering adopting, adding, or replacing in the next 12 months, 36 named Anthropic, a high ratio relative to its 45 current users. That suggests many businesses are still "trying" Claude without committing to it as core infrastructure.

Why this matters for small and mid-sized businesses: The gap between a platform a company tries and one it builds on is exactly the measurement missing from the ongoing argument about who's winning the enterprise AI market. When a business runs AI agents that take multiple steps — reading a file, calling a tool, deciding something, calling another tool — it needs a coordination layer that decides what runs when, what each agent can touch, and what happens when a step fails. That's the structural decision businesses are making right now, not just a question of which model is "smarter."

60% of surveyed enterprises plan to adopt, add, or replace an agent orchestration platform within 12 months. That's a moment when small and mid-sized businesses should examine not just which AI capabilities they're buying, but which infrastructure they're actually building their long-term business processes on.

RealOffice.AI is built around exactly this kind of choice as part of everyday operations: reliable AI agent infrastructure for the office that keeps supporting business processes once they move past the pilot stage and become part of daily operations.

Source: VentureBeat

Also in the News

Nvidia's CEO: '0% chance' AI ends the world. Jensen Huang told CBS Sunday Morning there's no chance AI could end humanity, calling Dario Amodei's (Anthropic) and Sam Altman's (OpenAI) calls to slow AI development "not grounded in science." The comment underscores how divided the industry remains on risk — notably, Huang's own net worth has climbed from an estimated $21 billion in 2023 to over $192 billion today. Source: The Verge

Startup designs chips with AI — in weeks instead of years. Architect Labs published research showing an AI system designed an inference chip (dubbed Redwood) from written spec to a working FPGA implementation with no human intervention below the spec level, in under two weeks. If the design holds up in fabricated silicon, it could significantly shorten the chip design cycle, which today is locked in years ahead of the models it will run. Source: VentureBeat