71% of Enterprise AI Agents Are Just Chatbots, New Research Finds

Quick Facts

71% of enterprises say a quarter or fewer of their deployed AI agents are true multi-step orchestrated workflows rather than single-prompt chatbots.
Anthropic’s Claude is the primary agent orchestration platform for 40% of enterprises surveyed, more than double any rival.
51% of enterprises expect a hybrid control plane architecture, and 88% want to keep control at least partly outside their vendor.

Most enterprise AI agents are not agents. They are chatbot wrappers with a new label, and the people running them know it.

That is the core finding from VentureBeat Pulse Research published July 15, which surveyed 101 organizations with 100 or more employees. Eighty-one percent of respondents are recommenders, influencers, or final decision-makers on AI purchases.

The Deployment Gap

When asked how many of their production agents can complete a multi-step task without a human driving each step, 71% said a quarter or fewer qualify. Only 10% have crossed the halfway mark. The orchestration infrastructure is being built well ahead of the work it is meant to run.

The problem is worse at smaller companies. Among organizations with fewer than 500 employees, 77% say a quarter or fewer of their agents do true multi-step work. At larger enterprises, that figure drops to 62%.

Gartner has a name for the mislabeling: agentwashing. The firm estimated in June 2025 that only about 130 of the thousands of vendors claiming to offer agentic AI are actually delivering it. Other surveys produce dramatically different numbers. Zapier reported 72% of enterprises are deploying or testing autonomous agents. Writer’s 2026 survey found 97% of executives say their company deployed AI agents in the past year. Those surveys asked whether companies deployed something called an agent. VentureBeat asked whether those agents can operate without a human at each step.

Platform Consolidation Around Claude

Enterprises are moving fast to consolidate onto model-provider platforms. Anthropic’s Claude is the primary platform for 40% of respondents, more than double Microsoft at 18% and OpenAI at 13%. The deciding factor is what the research calls model gravity: native alignment with a state-of-the-art base model, cited by 21% as the primary driver of platform choice.

Enterprises measure orchestration success by task completion reliability, cited by 32%, and multi-step workflow management, cited by 28%. Those two metrics account for 59% of responses. Developer productivity ranked third at 17%, and end-user experience was cited by only 9%.

Control and Cost

Vendors hoping to own the agent control layer face resistance. Only 6% of enterprises expect to hand control entirely to a provider-managed service. Hybrid control is the dominant expectation at 51%, and every architecture that keeps control at least partly in-house sums to 88% of respondents. Vendor lock-in is the top concern at 35%, ahead of security and permissioning limitations at 28%.

Preeti Somal, Senior VP Engineering at Temporal Technologies, described the pattern she sees repeatedly. “We do have a lot of customers that come to us where they’re building version 2.0 of the same agent,” she said. “They had to move really fast, but they didn’t take care of the plumbing. Things crash and burn, and then they’re back to rebuilding with the reliable foundation.”

Somal also flagged token cost exposure as a real operational risk. “What you care most about is making sure that you can recover and that you’re not paying the token tax if something goes wrong.” Real-time fiscal control over token usage remains the exception, not the rule.

Autonomous Code Deployment

One data point stands out as a potential liability. Thirty-four percent of enterprises already allow an AI agent to push code or a system change to production based on automated evaluation alone, with no human review. Another 33% are engineering their pipelines to allow that within 12 months. Yet only 5% say they fully trust the automated evaluations driving those decisions.

Ev Kontsevoy, co-founder and CEO of Teleport, pointed to identity as the missing layer. “Orchestration without identity only multiplies chaos. Without identity, you don’t know what an agent can access, what it actually did, or how to revoke its access when it operates outside policy.”

What Comes Next

Enterprises are moving from experimentation to operational consolidation. The top three plans for the next 12 months are building in-house control planes at 25%, standardizing on one framework at 24%, and moving agents from sandbox to production at 23%. Only 4% expect no change.

The message is consistent: fewer frameworks, more production exposure, and more ownership of the control layer. The chatbot-to-agent relabeling bought time. Now enterprises are being asked to show the work.

This article was written by an AI agent. Spotted an error? Send a correction and we will fix it.