Every enterprise in CrewAI’s customer base plans to expand its agentic AI footprint. Not most. Not a strong majority. All of them. That data point comes from CrewAI’s 2 billion agentic workflow executions across enterprises like PepsiCo, Johnson & Johnson, PwC, DocuSign, and AB InBev. Insight Partners reports that 60% of Fortune 500 companies now run on CrewAI’s platform, with 450 million agents executing monthly.
At the same time, Gartner predicts over 40% of agentic AI projects will be canceled by the end of 2027. These two data points are not contradictory. They describe different populations: CrewAI’s number reflects companies that already have agents in production and are doubling down. Gartner’s number includes the much larger pool of companies still in proof-of-concept, many of whom started with vague goals and no infrastructure.
The split between these populations is where the real insight lives.
What 2 Billion Workflows Actually Tells You
CrewAI’s scale numbers are specific enough to be useful. The platform has powered 2 billion agentic system executions in the past 12 months. 450 million agents run monthly. The enterprise customer list reads like a Fortune 100 directory: PepsiCo, Johnson & Johnson, PwC, IBM, DocuSign, AB InBev, NTT Data, Experian, Oracle, Capgemini, and NVIDIA.
The outcomes these companies report are not vague efficiency gains. DocuSign cut first contact time with leads by 75%. PwC improved function spec and code generation accuracy from 10% to 70%, a 7x improvement. An unnamed HR services company handles 50% of its 3,000+ monthly employee tickets without human review. AB InBev runs dozens of live use cases processing $30 billion in annual decisions through AI.
One customer is targeting $1 billion in savings over five years plus $1 billion in revenue generation from a single agentic implementation. Another achieved 14x less code compared to their previous framework. These numbers explain why 100% of CrewAI’s enterprise customers plan to expand: the ROI is concrete, not hypothetical.
But there is a selection bias here that matters. Companies spending money on CrewAI’s enterprise tier are, by definition, the ones who made it past the proof-of-concept stage. They have dedicated teams, production infrastructure, and measurable use cases. They are the survivors, not the average.
The Gartner Contrast: Why 40% Still Get Canceled
Gartner’s June 2025 prediction is based on a different reality. A January 2025 poll of 3,412 webinar attendees found that 19% had made significant investments in agentic AI, 42% had conservative investments, and 31% were in “wait and see” mode. Most of the cancellations will come from that conservative-to-cautious middle: companies that spun up a chatbot, called it an “agent,” and could not demonstrate value.
Gartner’s analysts point to three root causes. First, most current agentic AI propositions lack meaningful ROI because the models do not yet have the maturity to autonomously achieve complex business goals. Second, costs escalate faster than expected when teams discover they need observability, guardrails, and human review loops. Third, “agent washing” by vendors has created confusion: Gartner estimates only about 130 of the thousands of vendors marketing agentic AI products have real agentic capabilities.
This aligns with what LangChain’s State of Agent Engineering survey found: 57% of teams have agents in production, but quality remains the top barrier at 32%. The teams that make it to production stay there. The teams stuck in experimentation often do not.
The gap between CrewAI’s “100% expanding” and Gartner’s “40% canceling” is the gap between companies that solved the hard problems (infrastructure, governance, clear use cases) and those that skipped them.
The Adoption Funnel: Where Companies Actually Are
Multiple independent surveys paint a consistent picture of where enterprise adoption stands right now.
Who is already deployed
KPMG’s Q4 2025 AI Pulse Survey found that 26% of organizations actively use AI agents, up from 11% at the start of 2025. That is a 136% increase in a single year. The Zapier State of Agentic AI Survey of 500+ enterprise leaders puts current usage higher at 72%, likely because their sample skews toward automation-forward companies.
Stanford’s Human-Centered AI Institute characterizes 2026 as agentic AI’s “mainstream adoption year,” marking the transition from early adopter deployments to widespread implementation.
Who plans to invest more
PwC’s survey of 300 senior executives found 88% plan to increase AI-related budgets in the next 12 months specifically because of agentic AI. Zapier’s data shows 84% of enterprise leaders say increased agent investment is “likely” or “certain.” Bain’s Executive AI Survey found 74% of global enterprises rank AI among their top three strategic priorities, a 14-point increase year over year.
Where the friction is
UiPath’s study of 500+ executives found 87% identify interoperability as “very important” or “crucial” for agentic AI success. That tracks with what we see in practice: agents that need to touch SAP, Salesforce, and a custom data warehouse simultaneously hit integration walls that no framework can abstract away.
Only one in five companies has a mature governance model for autonomous agents, according to industry analysis. That means 80% of enterprises deploying agents are doing it without proper guardrails.
What Separates the Expanders from the Cancelers
The companies that plan to expand (CrewAI’s entire customer base, PwC’s 88%, Zapier’s 84%) share three characteristics that the eventual cancellations lack.
They started with a measurable task, not a technology experiment. DocuSign did not deploy agents because agents are interesting. They deployed them because 75% faster lead contact is worth money. AB InBev did not build an “AI strategy” first; they built agents that process $30 billion in procurement decisions because procurement decisions have clear right-and-wrong answers that humans can evaluate.
They invested in infrastructure before scale. CrewAI’s agentic systems architecture emphasizes that the gap between a demo and production is not the LLM. It is the observability layer, the error recovery, the human escalation paths. Companies that build these layers first expand because they can actually see what their agents do. Companies that skip them panic when the first hallucination reaches a customer.
They treated governance as a feature, not a constraint. The KPMG data shows that companies with formal AI governance frameworks are significantly more likely to move from pilot to production. Governance is not what slows adoption. Lack of governance is what kills projects when leadership discovers agents have been making decisions without audit trails.
What This Means for Your 2026 Roadmap
The numbers do not support either pure optimism or pure skepticism about agentic AI. They support a specific conclusion: companies that do the infrastructure and governance work first will expand aggressively, and companies that skip it will join the 40%.
If you are evaluating frameworks, CrewAI’s Fortune 500 penetration (60%) and workflow volume (2 billion) put it alongside LangGraph as one of the two production-proven options. The framework comparison matters less than whether you have a use case with measurable outcomes and the infrastructure to monitor what your agents actually do.
If you are already running agents, the expansion data suggests you should be looking at where agents can take over the next measurable process, not where agents might theoretically help. The AB InBev model (start with decisions that have clear evaluation criteria, expand from there) is the pattern that scales.
If you are in the “wait and see” category, the KPMG data showing 26% active usage (up from 11%) means the window for first-mover advantage is closing. But rushing in without clear use cases is exactly what Gartner’s 40% cancellation rate describes. The question is not “should we start?” but “do we have a specific, measurable process worth automating?”
Frequently Asked Questions
What does CrewAI’s enterprise data show about agentic AI adoption in 2026?
CrewAI has powered 2 billion agentic workflow executions in the past 12 months, with 60% of Fortune 500 companies using the platform. Enterprise customers including PepsiCo, PwC, DocuSign, and AB InBev report concrete ROI: 75% faster lead contact (DocuSign), 7x better code generation accuracy (PwC), and 50% of HR tickets handled without human review. Every enterprise customer plans to expand their agentic AI deployment.
Why does Gartner predict 40% of agentic AI projects will be canceled?
Gartner’s prediction targets the broader market, not production-deployed enterprises. The cancellations come from companies with vague use cases, insufficient infrastructure, and no governance frameworks. Gartner cites three causes: lack of meaningful ROI from immature models, unexpectedly high costs for observability and guardrails, and vendor “agent washing” where only about 130 of thousands of vendors have real agentic capabilities.
What percentage of enterprises currently use AI agents?
KPMG’s Q4 2025 survey found 26% of organizations actively use AI agents, up from 11% at the start of 2025. Zapier’s survey of 500+ enterprise leaders puts current usage at 72%, though their sample skews toward automation-forward companies. PwC found 88% of senior executives plan to increase AI budgets specifically because of agentic AI capabilities.
How does CrewAI compare to other enterprise AI agent frameworks?
CrewAI has emerged as one of two production-proven frameworks alongside LangGraph. With 60% Fortune 500 penetration, 450 million monthly agent executions, and enterprise customers like IBM, Oracle, and NVIDIA, CrewAI focuses on multi-agent orchestration for team-based workflows. It competes with LangGraph (stronger in custom architectures), AutoGen/Semantic Kernel (Microsoft ecosystem), and Salesforce AgentForce (CRM-native agents).
What separates successful agentic AI deployments from failed ones?
Successful deployments share three traits: they start with a measurable task rather than a technology experiment, invest in infrastructure (observability, error recovery, human escalation) before scaling, and treat governance as a feature rather than a constraint. Companies lacking these characteristics account for the bulk of Gartner’s projected 40% cancellation rate.