Agentic Reality Check: Only 11% of Companies Use AI Agents in Production
AI agents are everywhere right now. Vendors promise autonomous assistants that solve tasks on their own, make decisions on their own, scale on their own. Sounds great. But what’s actually happening?
Deloitte published its 17th annual Tech Trends report in December 2025, and one of the key chapters is titled plainly: “The Agentic Reality Check”. Authors Jim Rowan, Nitin Mittal, Parth Patwari, and Ed Burns compiled survey data, case studies from major companies, and expert opinions. Here’s what matters.
The Numbers: Gap Between Hype and Reality
According to Deloitte, the current state of AI agent adoption looks like this:
That means only 11% of companies are actually using AI agents in production. Meanwhile, 42% are developing a roadmap, and 35% don’t even have a formal strategy. Survey categories overlap - a company can pilot and develop strategy simultaneously - but the overall picture speaks for itself.
Gartner projections cited in the report:
- By 2028, 15% of daily work decisions will be made autonomously by AI agents (currently less than 1%)
- 33% of enterprise software will include agentic AI by 2028 (currently less than 1%)
- Over 40% of agentic projects will fail by 2027 due to legacy system incompatibility
Three Fundamental Problems
The authors identify three reasons why companies are stalling.
1. Legacy systems aren’t ready. Traditional enterprise systems were designed for humans, not agents. They lack real-time APIs, modular architecture, and secure identity management. Agents are forced to work through conventional APIs and data pipelines, creating bottlenecks.
2. Data isn’t structured for agents. Most companies’ data architecture is built around ETL processes and warehouses. According to the survey, 48% of companies cite data searchability as a problem for AI automation, 47% cite data reusability. Agents need data with business context, not raw tables.
3. No governance for autonomous decisions. Traditional IT governance models don’t account for systems making decisions on their own. This leads to “agent washing” - relabeling existing automation as “agents” - while poorly designed agents create “workslop,” making processes less efficient.
Don’t Pave the Cow Path
The report’s main thesis: companies are trying to automate processes designed for humans. That’s a dead end.
Brent Collins, VP AI Strategy at Intel, puts it directly:
Now is an ideal time for value stream mapping - to understand how processes should work versus the way they do work. Don’t simply pave the cow path. Take advantage of this AI evolution to reimagine how agents can best collaborate.
Ethan Mollick, Wharton professor and author of “Co-Intelligence,” adds:
It’s not actually a technology problem. It’s a process problem.
In my view, this is the key insight. I see the same pattern in banking: take an existing manual process with 15 steps and try to “automate” it step by step with an agent. Instead of asking - do we even need those 15 steps?
Who’s Doing It for Real
The report presents several case studies of companies that aren’t just deploying agents but redesigning processes.
Toyota: From 100 Mainframe Screens to One Agent
Toyota used 50-100 mainframe screens to track vehicle shipments. Much of the work was manual. Now a single agent provides real-time visibility from pre-manufacturing to dealer delivery. No mainframe interaction needed.
Jason Ballard, VP Digital Innovations at Toyota:
The agent can do all these things before the team member even comes in in the morning. We’ve made that critical decision to invest in this area. We feel like that’s where the differentiator is going to be going forward.
Next step: agents will independently identify shipment delays and draft resolution emails.
HPE: Four Specialized Agents Instead of One Universal
HPE built “Alfred” for operational reporting. Not one “smart” agent, but four specialized ones:
Marie Myers, CFO HPE:
We wanted to select an end-to-end process where we could truly transform rather than just solve for a single pain point.
Dell: A Dozen Pilots, but With Finance Sign-Off
Dell runs 12 agent pilot projects for composite tasks - quoting, end-to-end customer issue resolution across domains (entitlements, billing, logistics). 20 enterprise processes already digitized.
But there’s a strict rule: every agent requires material ROI sign-off from a finance partner and business unit head before going to production. An architectural review board evaluates and approves AI investments.
John Roese, CTO and Chief AI Officer at Dell:
AI is a process improvement technology. If you don’t have solid processes, you should not proceed.
To me, this is the key thought. Agents won’t save bad processes - they’ll scale them.
Moderna: HR + IT = One Function
Moderna went further organizationally. The company created the role of Chief People and Digital Technology Officer, merging HR and IT into one function.
Tracey Franklin, who took the position:
The HR organization does workforce planning really well, and the IT function does technology planning really well. We need to think about work planning, regardless of if it’s a person or a technology.
Multi-Agent Orchestration and Protocols
The report documents a shift from universal chatbots to specialized agents working together. Standards are already emerging:
Standard interface for connecting AI to enterprise data and tools. A universal connector between agents and enterprise resources.
Direct communication between agents across platforms. Agent discovery, task delegation, collaborative workflows.
RESTful API for agent interaction regardless of platform. An open alternative to proprietary solutions.
The approach resembles microservice architecture: many small specialized agents instead of one monolith. This simplifies debugging, testing, and scaling.
Hybrid Teams: Humans + “Silicon Workers”
The most provocative part of the report is the concept of a “silicon-based workforce.” Deloitte suggests treating agents as a new type of worker:
- Onboarding - agents need “training” on enterprise data, and people need training on working with agents
- Performance management - digital identifiers, cryptographic action receipts, immutable logs
- Lifecycle management - training updates, redeployment to priority tasks, decommissioning
Mapfre (insurance) already operates a hybrid model. Agents handle routine damage assessments, but sensitive customer communications stay with humans.
Maribel Solanas Gonzalez, CDO Mapfre:
It’s hybrid by design. With the high level of autonomy of these agents, it’s not going to substitute for people, but it’s going to change what they do today, allowing them to invest their time on more valuable work.
The Autonomy Spectrum
Deloitte identifies three phases:
The transition is managed through “agent supervisors” - humans who enter workflows at intentionally designed points to handle exceptions requiring human judgment.
What This Means
Deloitte paints a realistic picture. Agent technology works - Toyota, HPE, and Dell case studies prove it. But scaled adoption stalls for three reasons: legacy infrastructure, unstructured data, and missing governance models.
In my view, the main trap is “agent washing.” Relabeling existing automation as “agents” and reporting innovation. Real work starts with the question: which processes need to be redesigned from scratch, not just automated?
Dell requires ROI confirmation before deploying an agent to production. Moderna merged HR and IT to plan work - not people or technology separately. Toyota threw out 100 mainframe screens and rebuilt the process from scratch.
Most companies aren’t there yet. But the gap between hype and reality is exactly where the opportunity lies.
Source: Deloitte Tech Trends 2026, chapter “The Agentic Reality Check: Preparing for a Silicon-Based Workforce.” Authors: Jim Rowan, Nitin Mittal, Parth Patwari, Ed Burns. Projections: Gartner.
