Enterprise AI readiness reads 55.7 of 100, in the Operational band

This week the enterprise AI story split cleanly in two, and the fault line is not model quality — it is governed data.

On one side sits the acceleration. Deloitte reports worker AI access rose 50% in 2025 and that the cohort running 40%+ of experiments in production is on track to double within six months (MarketScale). Frontier vendors shipped the tooling to match: OpenAI Presence put an enterprise voice-and-chat agent platform into general availability (OpenAI), while Google expanded Managed Agents with remote MCP and background tasks for production-grade builds (Google). The product category is maturing on schedule.

On the other side sits the reckoning. MIT's Project NANDA found 95% of 300-plus enterprise deployments produced no measurable P&L impact, and Gartner records 50% of GenAI projects abandoned after proof-of-concept (Kanerika). The gap between the 88% who use GenAI and the 5% who profit from it is the defining statistic of 2026.

What separates the two sides? Read the wins. The fuel distributor that cut manual intervention 90%. The investment bank that hit 100% role-based compliance. FoodPharma unifying six systems on Fabric to compress reporting from two days to 90 minutes. Every documented win runs on a governed data layer — and Gartner confirms the inverse: 60% of AI projects without AI-ready data will be abandoned through 2026. The differentiator is not the model. It is the identity and data substrate the model stands on.

This is the Identity Control Surface thesis arriving in the numbers. IBM's IBV study found two-thirds of CIOs and CTOs are accountable for AI systems they do not fully control, and 70% say teams deploy faster than IT can track (IBM). Accountability without governed non-human identity is not governance — it is exposure. As agents move from copilots to autonomous actors (Gartner projects task-specific agents in 40% of enterprise apps by end-2026), the control gap becomes the value gap. You cannot capture ROI from decisions you cannot audit, from agents whose identity and permissions you cannot see.

The Janus Brands failure mode showed up too. Deloitte announced an Anthropic partnership the same week it was forced to refund an Australian department for a report riddled with AI hallucinations (Digiday). The AI message and the delivered artifact contradicted each other in public. Market AI Perception holds at 36 for exactly this reason: buyers have learned to discount the announcement and price the delivery.

The strategic read for this week: the leaders are pulling away, and the mechanism is not compute — it is the disciplined Compiled Corporation, where decisions run on governed data and every agent has a controllable identity. The median enterprise has AI access, pilots, and a CAIO. It does not yet have the substrate. Six dimensions moved up this week; every one of them was a product or infrastructure signal from vendors, not an outcome signal from adopters. That asymmetry is the story. The tools are ready. The organizations, mostly, are not.

Deloitte's finding that worker AI access rose 50% in 2025 is the clearest quantitative pull on this dimension in weeks. Access is broadening structurally, not episodically — reinforced by consumer surfaces like Google's connected-apps in Search AI Mode that normalize agent interaction inside daily tools. We move 65 → 66. This remains Systemic-tier: the distribution problem is largely solved; the depth-of-use problem is not.

The scaling picture is genuinely two-sided this week. Deloitte reports the share of firms with 40%+ of experiments in production is on track to double within six months — a real acceleration in the pilot-to-production chasm crossing. But Gartner's 50% POC abandonment rate and MIT NANDA's finding that 95% of deployments saw no measurable P&L hold the ceiling firm. The Compiled Corporation thesis is being tested in production, not slides. We move 61 → 62 — the leaders are scaling; the median is not. Restraint is warranted.

Governance frameworks are multiplying — but IBM's finding that two-thirds of tech CxOs are accountable for AI they do not control, and that 70% say teams deploy faster than IT can track, is a direct signal against runaway maturity claims. This is the Identity Control Surface problem in raw form: accountability without governed non-human identity is theater. AI Vanguard's operator note — governance that works is enforced at infrastructure, not policy — confirms the gap between documentation and control. We hold 77. The maturity is in the frameworks; the enforcement lags.

Deloitte reports that while 42% of companies say their AI strategy is highly prepared, confidence collapses when the question turns to talent. The strategy-execution gap is structural, and the CAIO trend — Fortune 500s appointing Chief AI Officers focused on business transformation and change management rather than model-building — signals recognition without resolution. No evidence justifies a move. Hold 63.

The ROI signal is bifurcated and both sides are cited from delivery data. Kanerika documents concrete wins — a fuel distributor's 90% reduction in manual intervention, 400+ hours saved monthly on governed data flows. Against that, MIT NANDA's 95%-no-P&L-impact finding across 300+ deployments is the dominant reality. The differentiator is explicit: value tracks to governed data. Where the data layer is disciplined, ROI appears; where it is not, deployment burns cash. We hold 60 — the winners prove the ceiling can be broken, but the median has not.

The CAIO wave — GM, Mastercard, Zocdoc, and agency holdcos appointing Chief AI Officers — is producing a new class of executives whose mandate is translating AI into business-impact language. This is messaging infrastructure, but it remains announcement-stage. No evidence of substantively differentiated AI-native brand narrative this week. Hold 44.

Google's expansion of Managed Agents in the Gemini API — background tasks and remote MCP — plus connected apps in Search AI Mode, moves the agent-callable infrastructure standard forward. MCP as a remote protocol is becoming the connective tissue for Decision Surfaces where agents call structured services. This raises the baseline expectation for what agent-ready infrastructure looks like. We nudge 51 → 52.

The perception gap holds. Deloitte's forced refund for a report containing AI-generated hallucinations — landing the same week as its Anthropic partnership announcement — is a textbook credibility puncture: AI ambition outpacing AI reliability in full public view. This is the Janus Brands failure mode — the AI message and the delivered artifact contradict each other. Market skepticism toward AI-competence claims remains justified. Hold 36.

Google Vids' Gemini Omni and personal avatars advance AI content creation, but this dimension concerns content structured for AI consumption — schema, machine-readability, retrievability. No direct evidence of enterprises restructuring content for agent ingestion this week. Evidence is tangential. Hold 35.

IPG's appointment of a global head of AI commerce and the broader agency CAIO trend signal intent to move brand/marketing operations toward AI orchestration. But these are leadership appointments, not evidence of AI-run marketing ops at scale. Hold 36.

OpenAI's launch of Presence — a proven enterprise agent platform for trusted voice and chat — pushes the interaction layer squarely into the agentic end of the spectrum for production customer and internal workflows. Combined with Health in ChatGPT connecting medical records for context-rich interaction, the assistant-to-agent transition is materializing in shipped products. We move 63 → 64.

Two frontier vendors shipped production-agent infrastructure this week: OpenAI Presence (enterprise voice/chat agents) and Google's Managed Agents with background tasks and remote MCP for reliable, production-ready builds. Gartner's projection that task-specific AI agents reach 40% of enterprise applications by end of 2026 (from under 5%) frames the trajectory. Microsoft Research's SkillOpt — treating agent skills as trainable parameters for reliability — reinforces the maturation. We move 67 → 68.

UX maturity — trust, explainability, error recovery — gets a modest research and product nudge. OpenAI positions Presence explicitly around trusted agents, and Microsoft Research's SkillOpt turns skill editing into a training process that makes agent behavior more reliable without changing model weights — a direct attack on the erratic-agent problem that erodes trust. Memora similarly tackles agent memory persistence. Reliability is the substrate of UX maturity. We move 55 → 56.

The FoodPharma case — six operational systems unified on Fabric, cross-functional reporting cut from two days to 90 minutes — is real process redesign, workflows rebuilt around a governed data spine rather than bolted-on automation. But Gartner's warning that 60% of AI projects lacking AI-ready data will be abandoned confirms most 'redesign' remains surface-level. The data foundation gates the redesign. Hold 53 — the exemplars are ahead of the field.

Value delivery mirrors ROI Impact: verifiable wins (43% faster retrieval, 100% role-based compliance) coexist with MIT NANDA's 95%-no-P&L finding. The distinguishing variable across every documented win is governed data feeding the AI feature. Where the substrate is disciplined, value is measurable and repeatable; elsewhere it evaporates. No net move justified. Hold 59.