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

This week the index splits cleanly along one fault line: deployment is outrunning control.

The deployment side is strong and, for once, well evidenced. The Brief Script froze a research core of 150 production agent deployments across 119 organizations, screened against a behavioral threshold and a verifiability protocol (The Brief Script). Their reading is the one we have been writing for a quarter: the shift is from human execution to machine execution inside explicit boundaries. Gartner's projection of agents in 40 percent of enterprise applications by year-end, up from under 5 percent, sets the slope (Agentic AI Institute). Presenc AI puts 78 percent of the Global 2000 with at least one production workload and a 2.4x median ROI (Presenc AI). Named cases filled in the pattern: Itaú ran an AI First strategy across customers and employees in one of Brazil's most regulated sectors (OpenAI), and Benefit Street Partners rebuilt credit-analysis follow-ups around Codex (OpenAI). This is the Compiled Corporation arriving in production: firms automating the repeatable decisions closest to their core.

Then the control side. 72 percent of enterprises run agentic AI in production without governance (Agentic AI Institute), and only 38 percent held a formal governance policy as of May (EverythingCloud). The governance guides converge on the same diagnosis: most programs have the policy half and almost none of the runtime half, so written standards and production drift apart (getmaxim.ai). That is the only dimension that fell this week, and it fell for a concrete reason. The Identity Control Surface, who the agent is, what it is allowed to touch, and whether that is enforced on every request, is the layer firms are shipping without.

The gap between the two sides is where the next incident cycle lives. An agent in production without an identity and a runtime control is not governed by its policy document. It is governed by whatever its credentials happen to permit. When 72 percent of deployments sit in that state, the governance score does not get to coast on the maturity of published frameworks.

There is a credibility footnote worth holding. IBM's self-reported 76 percent CAIO adoption sits against LinkedIn's verified 43 percent from actual title changes (Value Add VC). The inflation is instructive. The same optimism that reports a chief AI officer when the firm has a director reports governed agents when the firm has a policy PDF. Market perception stays in the Active band for exactly this reason.

The read for operators: the deployment curve is real, the return is real, and the control deficit is the binding constraint. The firms that will hold their scores are the ones installing identity and runtime enforcement on the request path before the agent count climbs further. Ship the control surface with the agent, not after the incident.

Access broadens as production workloads reach 78 percent of the Global 2000 (Presenc AI). Itaú's firm-wide ia.i interface shows access extending through employee and customer surfaces in a regulated sector (OpenAI). The move is up a point. Daily-workflow penetration is the gating question, and the evidence supports steady widening.

The scaling story firms up. The Brief Script's evidence-screened core of 150 production deployments across 119 organizations documents bounded autonomy doing real work (The Brief Script). Gartner's projection of agents in 40 percent of enterprise applications by year-end frames the trajectory (Agentic AI Institute). The move is up two points. The Compiled Corporation thesis sharpens: firms are moving execution inside explicit boundaries.

The governance dimension slips a point. The gap is now documented: 72 percent run agentic AI in production without governance (Agentic AI Institute) and only 38 percent hold a formal policy (EverythingCloud). Autonomy is outrunning the Identity Control Surface. Mature frameworks exist on paper, and the production drift between written policy and runtime enforcement is the live exposure. This dimension stays Systemic, but the evidence this week pulls it down.

Talent structures harden. Fortune 500 CAIO adoption sits near 43 percent on LinkedIn's title-change data, up from 19 percent a year earlier (Value Add VC). The role increasingly carries a change-management and literacy mandate, with Accenture's $3 billion hub as the scaled case. The move is up a point. Executive ownership is the precursor to workforce fluency programs.

ROI evidence strengthens. A 2.4x median return with software engineering at 2.9x and support automation at 3.4x gives the financial case named use-case detail (Presenc AI). Benefit Street Partners shows the gains landing inside credit-analysis workflows (OpenAI). The move is up a point. Returns are concentrating in support, code, and document analysis, the surfaces closest to repeatable decisions.

Limited direct evidence this week. Itaú's AI First framing, with ia.i and the IARA governance platform named as brand assets, is a clean Janus Brands example of legacy identity carrying a coherent AI narrative (OpenAI). Not enough to move the dimension. Holds flat.

The infrastructure layer for agents advances. Stack Overflow's agent-facing plugin tripled signups and drove up to 10x growth in agent creation (OpenAI), evidence that exposing machine-readable surfaces converts. NVIDIA's Nemotron edge and VL releases widen the model substrate for always-on agents (NVIDIA). The move is up a point. The Decision Surfaces question sharpens: brands that publish agent-readable interfaces capture agent traffic.

The credibility gap persists. The spread between IBM's self-reported 76 percent CAIO adoption and LinkedIn's verified 43 percent shows how self-reporting inflates AI competence claims (Value Add VC). This is the AI-washing dynamic that keeps market perception in the Active band. Holds flat.

Limited direct evidence. Stack Overflow's agent-targeted content property shows the commercial return on restructuring content for machine consumption (OpenAI). One case does not move the dimension; most enterprise content remains structured for human readers. Holds flat.

Marketing content generation is deployed at 33 percent of enterprises at 2.1x ROI (Presenc AI), the lowest-return use case in the set. Point-tool generation is not full orchestration of brand operations by AI. Evidence is tangential and holds the dimension flat in the Active band.

The interaction layer climbs. Itaú's ia.i puts a production conversational interface across customer and employee surfaces in a regulated bank (OpenAI). Gemini 4 Argon raises the model ceiling that assistant and agentic products build on (Google). The move is up a point as the spectrum shifts from chatbot toward agentic.

Agentic deployment leads the index. The Brief Script's frozen core of 150 production deployments documents bounded autonomy at scale (The Brief Script), and NVIDIA's Agent Toolkit binding 17 platforms lowers integration cost (Agentic AI Institute). The move is up a point. Production is the proof point, and the proof is accumulating.

UX maturity holds. Itaú's pairing of ia.i with the IARA governance platform is a trust-forward model (OpenAI), but the wider picture is 72 percent of agentic deployments running without governance (Agentic AI Institute), which erodes explainability and error recovery at the Decision Surface. The two signals offset. Holds flat.

Process redesign edges up. The Brief Script frames the real shift as moving execution inside explicit boundaries (The Brief Script), and Benefit Street Partners shows a credit-analysis workflow rebuilt around Codex (OpenAI). The move is up a point. Redesign is where the Compiled Corporation thesis gets tested, and bounded autonomy is the mechanism.

Value delivery firms up. Software engineering assistance posts 2.9x median ROI at 71 percent YoY growth (Presenc AI), and Stack Overflow's agent plugin tied a product feature to up to 10x growth (OpenAI). The move is up a point. Value is measurable where the AI feature sits on a repeatable workflow.