What's Changed Since Our Last Report

When we reported on enterprise AI agents failing at rates between 80-95%, the focus was on model limitations and unrealistic expectations. New data from May 2026 adds complexity to that story: the failure isn't just technical. It's organizational, strategic, and structural.

A global survey of 1,400 data analysts and IT leaders from Alteryx, published May 21, 2026 by ZDNet, found that 96% of IT professionals now use AI in their work. On the surface, that sounds like adoption success.

Only 49% use it frequently — meaning half the workforce is dabbling. For most organizations, this represents expensive experimentation rather than meaningful transformation.

The Unrestricted Access Problem

At least 50% of respondents said they are willing to grant AI agents unrestricted access to their data, according to the Alteryx survey reported by ZDNet.

The Alteryx survey didn't address the security implications. It noted, almost in passing, that 44% think human oversight should be part of that access. The other 56% apparently don't.

Half of IT professionals want to hand autonomous systems complete data access — with no agreed-upon safeguards — while those same systems are still failing at basic tasks like generating accurate recommendations.

What Agents Are Actually Doing

According to the Alteryx data via ZDNet, the top agentic AI use cases currently in production are:

• Drafting communications and summaries: 59%
• Scheduling and routing workflow tasks: 54%
• Generating standard reports: 48%
• Monitoring KPIs and triggering alerts: 45%
• Cleaning and validating data sets: 45%
• Running statistical analyses: 34%
• Automatically generating insights or recommendations: 23%

Only 23% of organizations are using agents for autonomous insight generation — the core function agents are designed for. Everyone else is using them for email drafting and calendar management. This is chatbot functionality dressed up as agentic AI.

The Time Tax

Data analysts are still spending close to six hours per week just prepping and validating data before AI can touch it, with 48% spending six to ten hours weekly on this work, per ZDNet's report of the Alteryx survey.

That's a full workday every week before a single agent runs. The AI efficiency promise carries a hidden labor cost rarely discussed in mainstream coverage. Validating AI outputs is becoming its own skill set. Automation has created a new manual job checking whether the automation worked.

New Infrastructure Is Rushing In Anyway

The market isn't waiting for enterprises to resolve these issues. On May 21, 2026, Kore.ai launched what it calls the Artemis edition of its Agent Platform — a direct challenge to Microsoft, Salesforce, Google, and ServiceNow in the enterprise AI infrastructure race, according to VentureBeat.

The technical centerpiece is the Agent Blueprint Language (ABL), a YAML-based compiled language designed to standardize how AI agents are defined, governed, and deployed. Kore.ai CEO Raj Koneru told VentureBeat the goal is to let businesses "design with AI, build with AI, test with AI, deploy with AI" — compressing months of engineering work into days.

ABL supports six built-in orchestration patterns — supervisor, delegation, handoff, fan-out, escalation, and agent-to-agent federation. It integrates with GitHub and standard CI/CD pipelines. On paper, it addresses the governance gap identified in previous analysis.

Whether enterprises will actually use the governance features remains unclear. Historical patterns suggest speed to deployment often wins out.

The Adoption-Strategy Gap Is Getting Worse

MIT Sloan Management Review research published in November 2025 — involving BCG's Shervin Khodabandeh among others — found that agentic AI has reached 35% enterprise adoption in just two years, with another 44% of organizations planning to deploy soon. Traditional AI took eight years to hit 72% adoption. Generative AI hit 70% in three years. Agentic AI is on pace to exceed both.

The MIT Sloan team's core finding: organizations are adopting agentic AI well before they have a strategy in place. A full 76% of executives in their survey now view AI agents as coworkers rather than tools — but very few have built management structures to match that thinking.

Deloitte's enterprise agentification report adds market context: 80% of automation leaders are expected to accelerate AI agent investments in 2025, the global agentic AI market is projected to hit $103.6 billion by 2032, and one-third of enterprise software applications are forecast to include agentic capabilities by 2028.

What's Being Overlooked

Most tech media covers this as a platform competition — Microsoft vs. Salesforce vs. Google. That framing obscures the actual challenge.

The bottleneck is NOT the models. It's NOT the platforms. Enterprises are deploying autonomous AI systems without governance frameworks, without clear data ownership, without trained staff to validate outputs, and — per the Alteryx data — with half their leadership willing to grant unrestricted data access to systems that haven't earned trust.

Appinventiv's May 2026 analysis identified the core traps: unclear strategy, poor data infrastructure, skills gaps, and companies stuck in "pilot purgatory" — running promising proofs of concept that never reach profitable production.

What Comes Next

Enterprise AI agents are being deployed faster than any previous technology wave. The infrastructure is improving — Kore.ai's Artemis launch represents real progress. But 96% adoption means little if agents are only drafting emails, data prep is still consuming a full day per week, and half of IT leaders are willing to grant autonomous systems unrestricted data access with minimal security review.

Faster deployment without better governance doesn't reduce failure rates. It scales them.