71% of enterprise AI agents are just chatbots, survey finds

Key Takeaways

• Only 10% of enterprises have crossed the halfway mark where most deployed agents are true multi-step workflows rather than chatbot wrappers
• Anthropic's Claude dominates enterprise agent orchestration at 40%, more than double Microsoft's 18% and OpenAI's 13%
• 27% of enterprises have no real-time way to stop runaway agent costs before the bill arrives

Enterprises have a labeling problem. A new VentureBeat Pulse survey of 101 organizations found that 71% admit a quarter or fewer of their deployed "agents" actually execute multi-step workflows autonomously. The rest are chatbot wrappers dressed up with a trendy name.

The findings expose a widening gap between orchestration ambition and deployment reality. Companies are building sophisticated control planes and investing heavily in agent infrastructure, but the agents themselves remain primitive. Only 10% of surveyed enterprises have crossed the halfway point where true orchestrated workflows outnumber single-prompt chatbots.

Who controls the agent platform market?

Anthropic's Claude Platform & Agent Skills leads by a wide margin, serving as the primary orchestration platform for 40% of respondents. Microsoft AI Foundry and Copilot Studio came second at 18%, followed by OpenAI's Agents SDK at 13%. Google's Enterprise Agent Platform captured 8%, Amazon Bedrock Agents 2%.

Open frameworks barely registered. LangChain and LangGraph accounted for just 6%. Another 5% built custom in-house systems, and 3% reported no orchestration at all.

The survey asked respondents to name their primary platform, so these figures represent where enterprises placed their main bet, not total spend across vendors. Still, the concentration is striking. Model gravity, the pull of a state-of-the-art base model, drove 21% of platform choices. Enterprises pick a model, then build around it.

What enterprises actually measure

When judging agent success, reliable multi-step execution dominates. Task completion reliability ranked first at 32% of respondents, followed by multi-step workflow management at 28%. These metrics reveal what enterprises expect agents to do: chain actions together without human intervention at each step.

But the metrics expose the gap. If task completion reliability matters most, and 71% of deployed agents cannot complete multi-step tasks, enterprises are measuring success criteria their agents cannot yet meet.

The vendor lock-in fear shapes architecture

Enterprises are not handing over the keys. By the end of 2026, 51% expect a hybrid control plane combining provider-native tools with external orchestration layers. Only 6% plan to rely entirely on a provider-managed service.

Vendor lock-in ranked as the top risk at 35% of respondents. The fear is justified: switching costs rise as workflows deepen their dependency on a single provider's model and API patterns. Enterprises want the model gravity without the gravitational collapse.

Investment priorities follow this logic. Agent workflow tooling leads spending at 34%, with security and permissions enforcement second at 25%. The money flows toward infrastructure that sits between the model and the production environment, layers that could theoretically swap providers.

Cost control remains primitive

Token costs add up fast when agents run autonomously. A poorly constrained agent can loop through API calls, burning through budgets before anyone notices. Yet more than a quarter of enterprises, 27%, have no real-time mechanism to stop a runaway agent before the invoice arrives.

This is not a platform problem. Workflow orchestration tools like n8n, Zapier, and Make have supported budget caps and execution limits for years. The gap suggests enterprises are bolting AI models onto existing infrastructure without the guardrails that automation tools long ago standardized.

Why the agent label sticks anyway

If most deployed agents are chatbots, why does the industry keep using the term? Three reasons.

First, internal politics. Calling something an agent signals ambition and secures budget. Chatbot budgets have been declining since 2020; agent budgets are growing.

Second, vendor marketing. Every major provider now sells agent platforms, not chatbot APIs. The terminology shift lets them charge enterprise pricing for capabilities that technically existed before.

Third, the roadmap. Enterprises are building the orchestration layer now, expecting the actual agents to mature later. The infrastructure bet is real, even if the deployed portfolio is not.

Methodology notes

The VentureBeat Pulse survey drew 101 respondents from organizations with 100+ employees in a single June 2026 wave. The sample skews toward technology and software (44%) and financial services (17%). Respondents were senior and purchase-adjacent: 81% influence or decide AI purchases. The sample is self-selected, not probability-based, so vendor share figures should be read as directional within this cohort rather than market-wide.

Frequently Asked Questions

What is the difference between an AI agent and a chatbot?

A chatbot responds to single prompts without autonomous action. An AI agent executes multi-step workflows, making decisions and calling tools without human input at each stage. The VentureBeat survey found 71% of deployed enterprise agents are actually chatbots.

Which platform leads enterprise AI agent orchestration?

Anthropic's Claude Platform & Agent Skills leads at 40% of enterprises surveyed, more than double Microsoft's 18% and OpenAI's 13%.

Why are enterprises choosing hybrid control planes for AI agents?

Vendor lock-in is the top concern at 35% of respondents. Hybrid architectures let enterprises use provider models while maintaining the ability to switch or add vendors later.

How do enterprises measure AI agent success?

Task completion reliability ranked first at 32%, followed by multi-step workflow management at 28%. Both metrics emphasize autonomous execution across multiple steps.

What is model gravity in AI platform selection?

Model gravity refers to choosing an orchestration platform based on the underlying AI model's capabilities. It was the top selection factor at 21%, meaning enterprises pick a model first, then build their agent infrastructure around it.

Source: AI | VentureBeat