The Production Problem

Everyone is deploying AI. Almost no one is getting value from it. This week, three separate institutions committed a combined $1 billion-plus to fix that -- and the data explains exactly why.

In this edition: EY + Microsoft's $1B "beyond experimentation" initiative  ·  NVIDIA/ServiceNow Project Arc  ·  Novo Nordisk's enterprise-wide OpenAI deployment  ·  Meta's 8,000 layoffs  ·  Intuit's 17% workforce cut  ·  AI-BOMs become mandatory  ·  The 97%/29% ROI gap  ·  Altman reverses on the jobs apocalypse

The $1 Billion Admission

EY and Microsoft just named the enterprise AI problem nobody wants to say out loud -- and put a billion dollars behind solving it.

The initiative combines Microsoft's Forward Deployed Engineers with EY's industry professionals across Finance, Tax, Risk, HR, and Supply Chain. The mechanism is deliberate: Microsoft provides engineering depth, EY provides change management and industry context. The underlying premise is that neither technology nor consulting alone can close the gap. You need both, working together on the same deployment.

EY's own rollout as "Client Zero" is instructive. The firm deployed Copilot to 150,000 users and reported a 15% productivity gain, which was reinvested into client delivery. That is not a pilot. That is a production deployment at meaningful scale. The 15% figure is notable not for its size, but for what it implies: productivity gains at that scale require organizational infrastructure, not just software installation. Someone built the change management program. Someone redesigned the workflows. Someone measured the outcomes and reinvested them.

This matters to enterprise leaders because it reframes the central question. The dominant AI conversation for the past three years has been about what AI can do. This week's $1 billion announcement signals that the dominant question for the next several years will be about how organizations operationalize it. The technology works. The execution, at most organizations, does not.

The evidence is uncomfortable. Across the enterprise AI landscape in 2026, 97% of executives report their companies have deployed AI agents -- but only 29% see significant ROI. That is not a margin of error. It is structural. The organizations inside the 29% are not using better technology. They are using the same technology differently: with tighter use-case selection, deeper workflow integration, better change management, and governance infrastructure built in from the start.

The harder question is what it means that institutional-scale intervention is apparently now required to cross the pilot-to-production threshold. After three-plus years of enterprise AI investment, two of the most sophisticated players in the industry are teaming up to solve a problem that should, theoretically, have been solvable by now. That it has not been suggests the gap is structural -- rooted in organizational design, incentive misalignment, and change management deficits that cannot be patched with more technology. POC Hell, as practitioners call it, is not a failure of imagination. It is a failure of execution infrastructure. And that infrastructure has now become a $1 billion market.

Moving Pieces

Five developments that matter to enterprise leaders this week

The Desktop Agent That Doesn't Quit: Project Arc

ServiceNow and NVIDIA unveiled Project Arc at ServiceNow Knowledge 2026 -- a persistent autonomous desktop agent for knowledge workers that operates continuously across multiple applications, completing multistep workflows without manual handoffs. Unlike cloud-based copilots that respond to single prompts then wait, Project Arc chains actions across tools and systems autonomously. What distinguishes it for enterprise adoption is the governance stack: NVIDIA OpenShell provides sandboxed, policy-governed execution, and ServiceNow's AI Control Tower logs every file read, command executed, and API called. For regulated industries, an agent that can be audited after the fact is a requirement, not a preference. Project Arc is currently in early preview.

Novo Nordisk Bets the Entire Enterprise on AI

Novo Nordisk's strategic partnership with OpenAI -- announced in April and gaining renewed attention this week -- deserves more coverage than it has gotten. The scope is enterprise-wide: drug discovery, clinical trials, manufacturing, supply chain, commercial operations, and workforce upskilling, all under a single AI deployment agreement with full integration targeted by end of 2026. What makes it notable is not the breadth, but the explicit treatment of workforce transformation as a core deliverable rather than an afterthought. Novo is not layering technology onto existing processes. It is redesigning workflows, with change management as a first-class component. For enterprise leaders watching from pharma and beyond, this is one of the clearest examples of what a full organizational commitment to AI production actually looks like.

Meta's $125 Billion Year Starts With 8,000 Layoffs

Meta began notifying roughly 8,000 employees on May 20 -- approximately 10% of its global workforce -- while simultaneously redirecting an estimated 7,000 additional employees into newly created AI-focused divisions, among them Applied AI Engineering and Agent Transformation Accelerator. Capital expenditure guidance for 2026 runs from $125 billion to $145 billion, more than twice the 2025 figure. Zuckerberg committed in an internal memo to no further company-wide cuts in 2026. The message is explicit: this is resource reallocation, not retrenchment. For enterprise leaders, the relevant question is not whether Meta's scale makes it exceptional -- it is whether a workforce restructuring model that pairs headcount reduction with explicit AI team expansion is becoming the standard playbook.

Intuit's 17% Cut and a CEO's Careful Framing

Intuit eliminated 3,000 positions -- 17% of its workforce -- on May 20, with restructuring costs estimated at $300 million to $340 million. CEO Sasan Goodarzi stated explicitly that the cuts "had nothing to do with AI." That framing landed the same week Intuit announced multi-year deals with Anthropic and OpenAI to embed their models into its tax and finance platforms. Whether AI caused the cuts is a question of attribution that may never be cleanly resolved. What is observable: Intuit, like Meta, is reducing headcount while expanding AI investment -- simultaneously. Organizations that name the connection between AI investment and workforce transition tend to manage it more effectively. The absence of a direct acknowledgment is itself information about how Intuit intends to handle the organizational change ahead.

AI Bills of Materials Are Becoming Mandatory, Not Optional

AI-BOMs -- inventories documenting the models, datasets, training history, licensing, and operational metadata behind AI systems -- are moving from optional security artifact to enforceable procurement requirement in 2026. The EU AI Act, coming into full effect in August, mandates documentation for high-risk AI systems. The FY26 National Defense Authorization Act requires AI component tracking for defense contractors. Cyber insurers are beginning to condition coverage on AI governance documentation, following the same underwriting playbook applied to ransomware risk in 2021. The gap between deployment and visibility is stark: 85% of organizations have AI in core operations, but only 25% have comprehensive visibility into how it is being used. For CIOs and CISOs, the AI-BOM is becoming the access ticket to regulated markets, government contracts, and institutional insurance coverage.

On the Radar

Five quick-hit developments worth tracking

of executives report significant ROI from AI deployments -- despite 97% saying their companies have deployed AI agents.

The 68-point gap between "deployed" and "value-generating" is the defining enterprise AI statistic of 2026. It is not a technology failure -- the models work. Survey data shows median payback periods ranging from 4.1 months for customer service deployments to 9.3 months for engineering, suggesting returns exist but are unevenly distributed and highly dependent on use-case selection and organizational readiness.

The $1 billion EY-Microsoft initiative announced this week is, among other things, a commercial bet on closing this gap systematically.

Sources: Writer.com 2026 Enterprise AI Adoption Survey  ·  Futurum Research

Altman Says He Was Wrong About White-Collar Jobs -- But the Data Is More Complicated Than That

Speaking in Sydney on May 26, OpenAI CEO Sam Altman told an audience that he was "pretty wrong" about AI's impact on white-collar employment. He had previously suggested entry-level professional roles were at serious near-term risk. Now: "I thought there would have been more impact on entry-level white-collar jobs being eliminated by now than has actually happened." The Yale Budget Lab supports the revision -- it finds no significant change in occupational mix or unemployment duration in high-AI-exposure roles since ChatGPT launched in late 2022. Anthropic CEO Dario Amodei, who once predicted 50% white-collar job elimination, has similarly retreated from that framing.

This revision arrives the same week Meta and Intuit collectively eliminated roughly 11,000 positions while expanding AI investment. The contradiction has a clean resolution: AI is not eliminating jobs at the macroeconomic level, but it is driving organizational restructuring at the firm level. Macroeconomic stability and individual displacement are not mutually exclusive outcomes -- and enterprise leaders managing workforce transitions should not confuse Altman's optimism for an argument against proactive change management.

Sources: Fortune  ·  Time  ·  4 Corner Resources

The Infrastructure of Execution

The word "experimentation" does a lot of work in enterprise AI conversations. It is the polite term for what practitioners call POC Hell.

This week, EY and Microsoft put $1 billion behind the premise that closing this gap requires institutional infrastructure, not just better technology or more ambitious roadmaps. The 97%/29% figure is worth sitting with. Ninety-seven percent deployment, twenty-nine percent ROI. That is not noise. The organizations inside the 29% are not using fundamentally different technology. They have something the majority do not: the organizational design to support production systems. Clearer use-case selection. Tighter workflow integration. Change management treated as a first-class deliverable. Governance infrastructure built before it was required, not after the audit.

The Novo Nordisk deployment and the Meta restructuring both gesture at the same truth from different directions. The companies getting value from AI are treating it as an organizational transformation, not a technology procurement decision. The companies stuck at 29% are still evaluating AI the way they evaluate software purchases -- a capability to license, install, and hope adopts itself. The production exit requires a different kind of commitment. Not more experimentation budget. Not another POC. A decision to build production systems and accept that the path from here to there will require rewriting some of what was built in the pilot phase.

Sam Altman's reversal on jobs is useful context, not reassurance. Macro stability and individual transition are different problems. Enterprise leaders managing workforces through this period are managing the latter. The aggregate data being stable does not make the organizational change management easier -- it just means the economy absorbs the displacement. Your organization still has to design the path through it.

The production exit is not a technology decision. It never was.

AK  /  Spearhead  /  Building AI systems that work