Enterprise AI readiness reads 61 of 100, in the Systemic band

This week the index moves on execution, not announcement. Every dimension that rose did so on named production evidence, and the pattern across them is one story told three ways: agentic systems are crossing from demonstrable to dependable, and the enterprises that redesign around them are booking returns the ones that bolt them on are not.

Start with the organization category. Presenc AI reports 78% of the Global 2000 running at least one AI workload in production, up from 41% two years prior, with a 2.4x median ROI. BeamSec calls the experimentation-to-operations shift complete. Read against the Compiled Corporation lens, this is the quiet part: firms are automating core operating steps function by function, with engineering and operations leading and HR and marketing trailing. The CAIO data from Value Add VC and ProfessionLens confirms AI leadership is now a standing executive seat, though the self-reported 76% figure is inflated by title fuzz and the verified Fortune 500 number sits closer to the low forties.

The governance line is where I hold firm. EverythingCloud reports only 38% of organizations have a formal AI governance policy, and 63% of firms that suffered an AI breach lacked one. NeuralTrust documents the specific failure: agentic patterns, multi-step action chains, cross-system tool access, memory persistence, outrun the frameworks built to govern them. Governance sits at the top of the organization category because policy intent is mature. It does not rise this week because runtime enforcement is not. This is the central tension in the Identity Control Surface framework: every agent deployed is a non-human identity that needs governing, and the NVIDIA/Microsoft push to run agents natively on Windows endpoints multiplies those identities faster than anyone is governing them.

The product category carried the most evidence. OpenAI's Sophos case puts a production agent on 52% of managed-detection cases with human oversight preserved, which is the Decision Surfaces pattern executed correctly: the human directs, the agent acts, the interface between them is designed rather than assumed. Oracle rebuilt specialist workflows across recruiting, engineering, and operations. LegalOn and Asana reframed value as cost at constant output. Microsoft Research's Agent Lightning lowers the cost of improving agents already in production. The gate on agentic deployment is now reliability engineering, not model capability.

The brand category stays stuck in the Active band, and that is the honest read. Agent-ready infrastructure rose on real endpoint and open-model evidence, but messaging, market perception, content readiness, and orchestration all held flat for lack of a clean catalyst. The Janus Brands question remains open: as AI messaging becomes an executive function, does it stay consistent with legacy identity or inflate into claims the product cannot back? The gap between IBM's 76% and LinkedIn's verified figure is the AI-washing risk made visible.

The read for operators: the capability gap has closed enough that your constraint is no longer the model. It is governance that enforces at runtime, workflows actually rebuilt, and identity controls for the agents you are already deploying. The firms pulling ahead treat agents as infrastructure under governance. The rest treat them as innovation projects, and the breach statistics are already telling us how that ends.

Access moves up a point. The Presenc AI figure of 78% of Global 2000 running production AI confirms workforce tooling is now standard in large firms, and BeamSec puts developer adoption near saturation. The Decision Surfaces read: the human/agent interface is now embedded in daily engineering and analysis work, no longer a sandbox. The constraint on further movement is breadth outside technical functions, where HR and marketing workflows still trail.

Scaling maturity rises a point. BeamSec states the shift from experimentation to operations is complete and the gap between early adopters and laggards is widening. Presenc AI anchors the production majority. Ecosystm adds a disciplining note: firms are pulling back from big-bet initiatives toward small-to-medium deployments with quarter-end outcomes, so scaling is broadening under tighter ROI gates. Against the Compiled Corporation lens, decision-making automation is proceeding function by function rather than wholesale.

Governance holds flat. The dimension sits at the top of the organization category, and this week's evidence argues against raising it. EverythingCloud reports only 38% of organizations have a formal policy, and NeuralTrust documents that agentic patterns outrun the frameworks meant to govern them. On the Identity Control Surface, non-human identity governance is where the gap is widest: the EU AI Act Article 50 transparency duties landed August 2026 while Annex III high-risk obligations slipped to December 2027 per Maxim AI. The score reflects mature policy intent, not yet mature runtime enforcement. Holding is the honest call.

Talent rises a point. The CAIO signal is the driver: Value Add VC reports the role nearly tripled year over year by IBM's measure, with LinkedIn's harder dataset showing Fortune 500 adoption roughly doubling. ProfessionLens confirms 24 clean enterprise-wide CAIO titles plus 54 fused titles among the Fortune 500. The honest caveat is that the IBM number is inflated by title fuzz. The move is real: AI leadership and literacy mandates are now a standing executive function, which pulls workforce fluency upward over time.

ROI moves up a point. Presenc AI quantifies a 2.4x median return with higher multiples in support, document analysis, and compliance monitoring. Ecosystm describes a sharper discipline: pilots without clear results are cut, so the returns that survive are better measured. Enterprise case evidence from OpenAI reinforces the pattern, with Sophos cutting threat investigation time by 96% and Oracle compressing specialist workflows. The score tracks documented financial impact rather than projected, which keeps it below the governance ceiling.

Messaging holds flat. The CAIO trend from Value Add VC shows AI communication becoming an executive function, which is a slow positive, not a this-week move. Against the Janus Brands lens, the open question remains whether enterprise AI messaging stays consistent with legacy brand identity or drifts into claim inflation. Evidence this week is tangential, so the prior score holds.

Infrastructure rises a point. The NVIDIA/Microsoft RTX Spark announcement pushes agent execution to Windows endpoints, and the telecom open-models report shows operators building for control and customization across autonomous networks. On the Identity Control Surface, endpoint agents multiply the number of non-human identities that need governing, which is a structural shift in where agent-ready infrastructure must sit. The move is modest because governance of these new surfaces lags the capability.

Perception holds flat. The dominant signal cuts both ways: Value Add VC documents real CAIO acceleration alongside clear overstatement in self-reported surveys. That tension is exactly what keeps market perception in the Active band, where credibility and AI-washing coexist. No clean positive catalyst this week, so the score holds.

Content readiness holds flat. Direct evidence on content structured for AI consumption is thin this week. The NVIDIA open-models signal is tangential: controllable infrastructure is a precondition for machine-readable content, not proof of it. Holding the prior score is the honest call given limited evidence.

Orchestration holds flat. OpenAI's Pollo AI case shows campaign production moving to model orchestration, but this is a creator-tool story rather than evidence of enterprise brand operations run by AI. The signal supports the direction of travel without justifying a this-week move. Prior score holds.

The interaction layer rises a point. OpenAI's Asana case shows browser agents reaching viable cost and latency, and the NVIDIA/Microsoft Windows integration moves agentic interaction toward a platform default. The spectrum is tilting from chat toward agent-driven action inside the product surface. The Decision Surfaces read: the human now directs agents rather than operating tools, which is the defining shift in this dimension.

Agentic deployment rises a point, the strongest-evidenced move in the product category. OpenAI's Sophos case puts a production agent on 52% of managed-detection cases with oversight preserved, which is the Decision Surfaces pattern done well. CXOTalk confirms 1,000-plus agent fleets at the top tier, and Microsoft Research's Agent Lightning lowers the cost of improving agents already in production. Deployment is broadening structurally, with the gate now reliability engineering rather than model capability.

UX maturity holds flat. The OpenAI Sophos oversight model is a positive signal on error recovery, but Microsoft Research documents that surprising failures in complex settings remain unsolved. Trust, explainability, and recovery are advancing in pockets without a clean step-change this week. Holding the prior score is the honest read.

Process redesign rises a point. OpenAI's Oracle case shows specialist workflows reconstituted as repeatable AI-run processes across three functions, and BeamSec documents Doctolib replacing legacy testing infrastructure outright. This is the Compiled Corporation pattern: core operating steps rebuilt around agents rather than decorated with them. The move is modest because redesign remains concentrated in engineering and operations.

Value delivery rises a point. OpenAI's LegalOn and Asana cases show value measured as cost-at-constant-output, a sign the metric is maturing beyond raw capability claims. Presenc AI anchors the financial picture with engineering ROI near 2.9x and the fastest year-over-year growth of any use case. Value delivery is becoming quantifiable and defensible, which is what moves this dimension.