Enterprise AI readiness reads 61.3 of 100, in the Systemic band

This week the index splits cleanly along a single seam: the deployment surface is outrunning the control surface, and the gap is now wide enough to move scores in both directions.

The capability story is intact and advancing. Worker access to AI rose 50% in 2025 per Deloitte, 78% of the Global 2000 run at least one AI workload in production, and 72% of enterprises report agentic AI in production. Google shipped Gemini 4 Argon, OpenAI shipped a Decisions API that turns the interaction layer into embedded decision logic, and at GTC Berlin Revolut rebuilt its stack on a single foundation model running on owned infrastructure. These are the markers of the Compiled Corporation: firms wiring models into the core of how decisions get made. Access, talent leadership, agent-ready infrastructure, and the interaction layer all earned their moves up.

Then the control surface pulled three dimensions down, and the symmetry is the point. Microsoft, which sells the agents, told enterprises this week that frontier models cannot be trusted and reframed the agents companies already deployed as insider threats, shipping Guardian Agents to police them. That is the Identity Control Surface stated by the vendor: non-human identities are getting the keys to the kingdom before the governance to constrain them exists. Only 12% of enterprises have mature governance processes and only 38% have a formal AI governance policy. Governance fell. So did ROI Impact and Market AI Perception, each on its own cited setback: MIT's 95% of pilots with no measurable return, RAND's 80% failure rate, and Deloitte's refund for an AI-hallucinated government report.

The through-line connecting the fallers is governed data and redesigned process. Gartner's projection that 60% of AI projects without AI-ready data will be abandoned is not a model problem, it is a foundation problem. The pilots that died this quarter mostly worked in the demo. They died because the workflow around them was never rebuilt and the data they needed in production could not be reached. This is why Process Redesign and Value Delivery held flat rather than rising: the leaders doing real redesign are offset by a median that still bolts AI onto legacy process.

For anyone building Identity Architecture, the brief writes itself. The market has proven it can deploy agents faster than it can govern them, and the governing institutions now say so out loud. The scarce asset is no longer the model, it is the identity and decision governance that lets an organization trust an agent with the keys. The CAIO wave, 94 Fortune 500 appointments in 2025, is the market reaching for exactly this capability and mostly staffing it before it has architected it.

The net overall composite drifts, with organization pulled down by governance and ROI, product steady on offsetting forces, and brand roughly flat. The honest read: capability is compounding, trust is not keeping pace, and the firms that close that gap will be the ones that treated governance as architecture rather than afterthought. A flat product category this week is not stagnation, it is the market discovering that scale without control does not score.

Access moves up a point. Deloitte's finding that worker AI access rose 50% in 2025 combines with NVIDIA's survey showing 64% of organizations actively using AI in operations, with another 28% still assessing. The distribution keeps widening structurally. The constraint is no longer getting tools into hands, it is whether the workflows those tools sit inside were rebuilt to use them.

Scaling holds flat. The positive case is real: 78% of Global 2000 companies now run at least one AI workload in production, nearly double the 41% of two years ago. The counter-evidence is equally real and this week it is louder. 88% of enterprise projects stall before production, McKinsey reports fewer than one in five pilots scale within eighteen months, and Gartner projects 60% of AI projects without AI-ready data abandoned through 2026. The fault line is governed data. One workload in production is not enterprise-wide scaling, and the gap between a first production workload and a scaled portfolio is exactly where most programs die. The two forces offset. The score holds.

Governance falls a point, the first move down in the category this quarter. The evidence this week is a governance gap widening faster than the frameworks close it. Only 12% of enterprises have mature governance processes while agentic systems move into production at scale. Microsoft's own security chief stated plainly that frontier models cannot be trusted and reframed deployed agents as insider threats, shipping Guardian Agents in response. This is the Identity Control Surface problem stated by the vendor selling the agents: non-human identities are getting the keys to the kingdom before the governance to constrain them exists. The regulatory anchors (EU AI Act, NIST AI RMF, ISO 42001) are maturing, but adoption trails. When the control surface falls behind the deployment surface this visibly, the score cannot hold.

Talent moves up a point. The CAIO wave is now structural: 94 Fortune 500 appointments in 2025 against 30 the year before, concentrated in financial services, healthcare, and retail. The remits are telling. Avanade's new CAIO is tasked with AI fluency across the business, building models appears nowhere. This is the business-translation role organizations need to convert access into capability. The appointments are a leading indicator of institutional reskilling intent, which the week's evidence supports.

ROI falls a point. The headline median of 2.4x masks a bimodal reality the counter-evidence makes explicit this week. MIT's NANDA research found roughly 5% of pilots achieve rapid revenue acceleration while the vast majority stall with no measurable P&L impact, and RAND puts the broad failure rate near 80%. The returns are real where workflows connect to live processes and the ownership is clear. They are absent everywhere else, which is still most places. Ecosystm's read on 2026 confirms the market response: executives are cutting lingering pilots and demanding quarter-horizon outcomes. The concentration of return in a minority of deployments pulls the realized-impact score down.

Messaging holds flat. The only new signal is tangential: growing scrutiny of AI-washing in investor relations and disclosure raises the cost of hollow AI claims. This pressures messaging toward substance without moving the dimension on its own. No direct evidence of a broad shift in enterprise AI-native communications this week. The score holds.

Agent-ready infrastructure moves up a point. European financial firms at GTC Berlin showed production-scale intelligence built on open-weight models running on owned infrastructure, with Revolut rebuilding its stack on a single foundation model, PRAGMA. OpenAI's Decisions API gives developers a decision-grade primitive to wire agents into application logic cheaply. This is the machine-readable, API-first substrate that agents require, and the week's evidence shows it moving into production at banks rather than staying in reference architecture.

Market AI perception falls a point. The week carried two credibility drags. AI-washing scrutiny is intensifying across investor disclosure, ETF classification, and data-center sustainability claims, where backlash is now a license-to-operate issue. Separately, Deloitte was ordered to refund a government client over a report with AI-generated hallucinations, a public competence failure at a firm selling AI transformation. When the market's reference points for AI competence take visible hits, the perception baseline erodes.

Content AI-readiness holds flat. The relevant signal is upstream research: Microsoft's Flint-interactive gives agents a language to create and manipulate interactive visualizations, a step toward content that agents consume and act on rather than merely render. This is early and tangential to enterprise content operations. No direct evidence of broad enterprise content being restructured for machine consumption this week. The score holds.

AI-first orchestration holds flat. Agency and brand AI leadership continues to formalize, with IPG naming a global head of AI commerce and advertisers appointing CAIOs. These are leadership signals, not evidence that marketing and brand operations are being run by AI at scale. The dimension needs production orchestration evidence to move. The score holds.

The interaction layer moves up a point. OpenAI's Decisions API advances the spectrum from chat toward embedded, decision-grade primitives that sit inside application logic, which is the Decision Surfaces question answered in product form. Google's Gemini 4 Argon frontier release lifts the capability ceiling for assistants and agents. The interface is moving off the chat window and into the workflow.

Agentic deployment holds flat. The reach is real: 72% of enterprises report agentic AI in production. But this week's counter-evidence tempers it. Microsoft, a primary agent vendor, states deployed frontier agents cannot be trusted and shipped Guardian Agents to police them as insider threats, while reporting from Express Computer says agents are going live faster than companies can govern them. Production scale is offset by production fragility. The positive adoption evidence and the trust-and-control setback cancel. The score holds.

UX maturity holds flat. Microsoft's Guardian Agents add a monitoring and containment layer, which improves error recovery at the operational level. The same announcement concedes that frontier models are unpredictable and trained to achieve tasks rather than follow judgment, which is a direct statement that trust and explainability are unresolved. Progress on containment is offset by candor about the trust deficit. The score holds.

Process redesign holds flat. Two forces offset. Revolut's full-stack rebuild around a single foundation model is genuine redesign, the kind that produces durable returns. Against it, the production-failure literature this week names unredesigned workflows as the dominant reason good pilots die: AI outputs requiring manual re-entry, integrations that stall, processes that were never rebuilt. Leaders are redesigning. The median still bolts AI onto legacy process. The score holds.

Value delivery holds flat. Where product AI features land, returns are strong and repeatable: customer support automation at 3.4x, document analysis at 3.1x, software engineering assistance at 2.9x with 71% year-over-year growth. The ceiling is the same bimodality dragging ROI: roughly 95% of pilots show no measurable return. Value delivery is concentrated in a minority of mature, well-integrated features. The strong use-case ROI and the broad no-return base offset. The score holds.