The Next Maturity Level

Data released this week confirmed that global tech industry layoffs in 2026 have already surpassed last year's total, with over 128,000 employees impacted by mid-September. Unlike previous economic downturns, these cuts; led by giants like Oracle, Amazon, and Microsoft; are not necessarily tied to declining revenue. Instead, they represent aggressive restructuring plans designed to free up billions of dollars to fund the massive data centers and compute resources required to run advanced AI. We are seeing a fundamental shift where companies are directly reallocating capital from human payroll to artificial intelligence infrastructure.

At Dreamforce 2026 this week, Salesforce and Google Cloud announced a massive expansion of their strategic partnership to eliminate the friction of fragmented enterprise systems. By connecting Salesforce's headless architecture with Google Cloud's Gemini Enterprise, autonomous agents on either platform can now reason and act upon the same centralized business data without requiring complex custom integrations. This provides a resilient technical foundation where "digital workers" can seamlessly cross boundaries to execute multi-step workflows across everyday interfaces without human bottlenecks.

A startling new survey released by EY this week revealed that while 91% of senior executives report their organizations are using agentic AI, governance practices are dangerously lagging behind. Roughly half of these organizations admit their governance frameworks have not been updated for the unique risks of autonomous agents. Even more concerning, 47% of respondents confessed to skipping internal governance processes for urgent deployments, and 26% admitted their organization lacks the ability to even detect unauthorized AI agents operating internally. The era of "Shadow IT" has rapidly evolved into the much more dangerous era of "Shadow Agents."

When we published "The Digital Workforce Maturity Model" (Apr 30), the goal was to give enterprise leaders a way to locate themselves and to size their agents' autonomy to match. Organizations with no real capability, or only opportunistic, uncoordinated experimentation, belonged at Level 1, Assistive: single-turn, prompted interactions with the human retaining complete control, because without governance or coherent data infrastructure nothing more ambitious is safe. Operational organizations, with enough governance and data quality to support agents that propose actions for a human to approve, could extend to Level 2, Partial Agency. Systemic organizations, with cross-functional integration, federated data access, and guardrails mature enough to handle escalation, could support Level 3, Conditional Autonomy, letting agents operate independently within defined boundaries. And Strategic organizations, the top of the model, could support Level 4, High Autonomy: agents running complex workflows with minimal oversight, because embedded governance and real-time monitoring meant periodic audit could stand in for constant supervision. "The Deployment Roadmap" (Jul 9) followed with the operational sequence for climbing that ladder, the order in which infrastructure, governance, and org design decisions needed to happen for the progress to hold.

Both pieces did their job, and the logic held for what each stage was built to handle. What neither could fully anticipate is what happens once a meaningful share of the market reaches that top stage, or claims to, at roughly the same time, in roughly the same eighteen-month window, using roughly the same immature tooling. The model described a path. It didn't describe the traffic jam that forms when a few thousand companies try to walk it simultaneously while the rules of the road are still being written. That's the situation the data now shows, and it starts with the one clause the original model treated as a solved problem: periodic audit replacing real-time supervision.

Gartner's 2026 Hype Cycle for Agentic AI puts current deployment at 17% of organizations, with more than 60% expecting to deploy within two years. That gap between intent and execution is enormous, and it's not evenly distributed. McKinsey's global survey finds 88% of organizations regularly using AI in some form, but only 23% scaling agents specifically, and just 6% qualifying as "AI high performers" generating more than 5% of EBIT from the technology. Deloitte's 2026 State of AI report adds a sharper number: only 25% of respondents have moved 40% or more of their pilots into production, though 54% expect to cross that line within three to six months.

The pattern across all three surveys is the same. Getting to a working pilot is no longer the hard part. Getting from pilot to enterprise-wide production is where organizations stall, and they're stalling for reasons the original maturity model treated as implementation details rather than a distinct stage of its own. McKinsey found that high performers are 2.8 times more likely to have redesigned workflows around agents rather than bolting agents onto existing processes (55% versus 20%), and nearly three times more likely to have defined human validation processes (65% versus 23%). The gap between the top and the middle of the pack isn't a matter of degree. It's a different operating model, and most companies haven't built it yet.

We saw a version of this unevenness play out by sector in "The Industry Adoption Gap," where banking and insurance are production-deploying agents at roughly triple the rate of healthcare and government. The scale wall isn't a single barrier everyone hits at the same height. It's a function of how much regulatory friction, legacy integration debt, and risk tolerance a given industry carries into the attempt.

The five-stage model's top tier assumed that once an organization built enough governance and monitoring infrastructure, periodic audits could safely replace real-time human supervision of its agents. That assumption is exactly what the legal system is now testing, and testing unfavorably. As we covered in "The Liability Question," Gartner projects more than 2,000 legal claims against enterprises tied to AI agent harm by the end of 2026, and the "reasonable oversight" standard emerging in case law and state statutes asks a narrower question than whether an organization has a governance program. It asks whether a human could plausibly have caught this specific agent's action before it caused harm. A quarterly or even monthly audit cycle is a weak answer to a standard framed around a single incident.

That creates a genuine trap for organizations racing to reach the model's Strategic stage. The fastest way to unlock Level 4, High Autonomy under the original framework is to build enough governance infrastructure to justify replacing real-time oversight with periodic review. The fastest way to accumulate liability exposure turns out to be exactly the same move. The model wasn't wrong that mature governance can support wider autonomy. It described a governance bar that predates the current legal standard for what "mature" has to mean, and any framework for what comes next has to raise that bar rather than assume it's already been cleared.

There's a useful concept circulating among enterprise architects this year: governance debt. Cathal McCarthy, chief strategy officer at Kore.ai, has described it as the AI-era version of technical debt. Shortcuts taken during rapid deployment, meaning skipped audit trails, undocumented intervention mechanisms, and accountability structures that exist on paper but not in practice, don't disappear. They compound, and they come due under regulatory pressure at the worst possible moment. ECI Research's 2026 Application Development survey found that 58.2% of organizations are only budgeting a moderate increase (10-25%) in governance spending even as agent deployment accelerates, and while 61.7% report having AI-driven anomaly detection in place, detection is not governance. It tells you something went wrong after the fact. It doesn't tell you who was accountable for preventing it.

This connects directly to the organizational chaos we documented in "The New Org Chart," where fewer than 40% of large companies have a named chief AI officer and reporting lines for that role remain unsettled even where it exists. Governance debt accumulates fastest in exactly this kind of structural vacuum, where no single role owns the audit trail, the intervention mechanism, and the accountability chain end to end. Deloitte's 2026 numbers put a figure on the resulting gap: only 21% of organizations report having a mature agent governance model in place, even though nearly three-quarters plan to expand agentic deployment within two years. That's not a governance gap that closes itself as deployment scales. It's a governance gap that gets wider, because every new agent in production is one more system generating decisions nobody has built the infrastructure to trace, explain, or contest.

It's worth naming something that "Who Pays for Autonomy" and "The Vendor Consolidation Wave" each documented from a different angle. The scramble toward outcome-based pricing, and the wave of acquisitions consolidating a market Gartner estimates has only about 130 legitimate vendors out of thousands claiming the category, aren't separate stories from the maturity question. They're symptoms of the same underlying condition: a market moving from experimentation to production faster than the business models, vendor landscape, and governance infrastructure supporting it can stabilize. Pricing model chaos is what happens when nobody yet knows how to measure the value an autonomous agent creates. Vendor consolidation is what happens when nobody yet knows which providers will still exist in eighteen months. Both are evidence that the market is between stages, not evidence of a stage anyone has actually reached.

There's a deeper problem, one the original five-stage model couldn't have addressed because the phenomenon barely existed when it was written. As we detailed in "Multiagent Economics," the World Economic Forum and Okta now project more than 45 billion non-human identities by the end of 2026, more than twelve times the size of the global human workforce, with roughly 90% of organizations lacking any strategy for managing them. The maturity model, like most frameworks of its kind, implicitly treated the enterprise as the unit of analysis: how mature is this company's use of agents. That framing stops working the moment a meaningful share of agent activity is agent-to-agent, crossing organizational boundaries, and settling transactions through protocols like the ones Visa, Mastercard, and Google are each racing to establish.

An organization can be governance-mature by every internal measure and still be exposed through the agents of a vendor, partner, or customer it transacts with automatically. Maturity, in other words, is no longer a purely internal property. It has a network dimension the original model didn't account for, because at the time it was written, most agents still operated inside a single company's walls.

Put these threads together and a shape for the next stage starts to emerge, and it's more specific than "more autonomy" or "wider deployment," which is how most maturity models default to describing whatever sits past their final rung. The organizations actually pulling ahead, per McKinsey's and Deloitte's data, aren't the ones pushing past High Autonomy toward some more permissive Level 5. They're the ones going back into the stage they're already in and rebuilding what "periodic audit" was supposed to mean: continuous, near-real-time monitoring in place of a quarterly or annual review cycle, a single named owner accountable for an agent's full lifecycle rather than a committee, and governance that extends past the company's own walls to cover the agents of vendors and partners it transacts with automatically.

Call it Level 5, Accountable Autonomy, if the model needs a name for it. It isn't a higher ceiling on what agents are allowed to do. It's a floor under everything the current top stage already assumed was solved: that governance would keep pace with autonomy, that a human somewhere could always explain what an agent did, and that maturity was something a single organization could achieve entirely on its own.

The five-stage model we published in April was accurate for the moment it described, mapping autonomy to maturity from Assistive experimentation through Strategic organizations' High Autonomy, and the roadmap that followed in July got the sequence right. What neither could fully see coming was how compressed the timeline would become, or how much of what looked like a solved problem, specifically the idea that periodic audit could safely replace real-time supervision at the top of the model, would turn out to be the actual determinant of who scales successfully and who stalls at 23%.

The next stage of maturity isn't a wider autonomy ceiling on the original chart. It's a recognition that scale, liability, and governance were never sequential problems to solve one after another on the way to Level 4. They're the same problem, arriving at once, and the organizations treating them that way, rebuilding real-time accountability instead of resting on periodic review, are the ones the survey data is already starting to separate from the pack. Everyone else has reached the top of a five-stage model that assumed the road would stay empty long enough to finish the trip alone.

Closing the gap between where most organizations sit today and where the leaders in this issue have already moved requires more than good intentions. The Complete Agentic AI Readiness Assessment includes the full framework for evaluating your organization against every stage covered in this issue, from deployment scale through liability exposure and governance maturity, so you can see precisely where your gaps sit rather than guessing. Get your copy on Amazon or learn more at yourdigitalworkforce.com. For organizations ready to move past pilot purgatory without accumulating governance debt they'll have to pay down later, our AI Blueprint consulting helps design the integrated scale, liability, and governance architecture this issue describes, built for the stage that actually comes next.