The Transformer's Signal — Edition 07

14 April 2026

This Week's Signal

McKinsey published a 28-page report confirming that nearly everything enterprises have done with generative AI has produced no measurable business result.

The report, "Seizing the Agentic AI Advantage" (QuantumBlack, June 2025, fully analysed this cycle), introduces what McKinsey calls the "Gen AI Paradox": 78% of companies have deployed generative AI in at least one business function. Roughly the same percentage report no material contribution to earnings. Only 1% view their strategies as mature.

The statistic that should concern every leader in mid-flight: fewer than 10% of vertical (function-specific) use cases ever make it past the pilot stage. Ninety percent of the use cases with real economic potential die before reaching production.

McKinsey identifies a three-level adoption model. Level 1: bolt AI onto existing workflows (5-10% gains). Level 2: AI assists within workflows (20-40%). Level 3: reinvent processes around agents as participants (60-90%). Almost every enterprise is stuck at Level 1. The chasm between Level 1 and Level 3 is not technological. It is organisational.

Their conclusion: "The bigger challenge won't be technical. It will be human: earning trust, driving adoption, and establishing the right governance."

We explored the full implications in this week's Way Pointer deep dive: The Gen AI Paradox — why 78% deployed and 80% have nothing to show for it.

Three Things Worth Knowing

Marc Andreessen told the Latent Space podcast that "entire industries are going to get built" to help organisations absorb AI. In a 76-minute conversation (April 3, 2026), Andreessen argued that the binding constraint on AI is not model capability but institutional absorption capacity. Using James Burnham's framework, he described managerial capitalism as structurally unable to absorb this pace of change. The technology investor who backed Netscape and GitHub is saying what McKinsey's data proves: the adaptation layer is missing, and building it is the next great industry.

A peer-reviewed study found that experienced developers using AI coding tools performed 19% slower — while believing they were 24% faster. The METR study, cited this month by Nate B Jones, reveals a 43-percentage-point gap between perceived and actual AI productivity. This is what happens when organisations bolt AI onto existing workflows without restructuring the work itself. The J-curve applies beyond software: any team adopting AI tools without operating model redesign is likely getting slower while reporting gains.

Ramp's April 2026 AI Index shows 50.4% of businesses now pay for AI tools — up from 35% a year ago. The adoption majority threshold has been crossed. Anthropic's market share reached 30.6% versus OpenAI's 35.2%, narrowing rapidly. Meanwhile, Samsung posted $37.9 billion in Q1 operating profit (8x year-on-year) on surging AI memory demand, and CoreWeave's backlog hit $90 billion. The infrastructure economics are shifting: AI pricing subsidies are ending. Prepare for significant cost increases.

The Pattern

This week produced the densest convergence of evidence we have tracked. McKinsey, Andreessen, a peer-reviewed productivity study, and enterprise spending data all point to the same structural truth: the technology works, but the organisations deploying it do not change fast enough to capture the value.

The Gen AI Paradox is not new. We have been tracking this since NBER's 6,000-executive study in February. What is new is the weight of authority behind it. When the world's most-cited strategy consultancy publishes a 28-page report reaching the same conclusion as the world's most influential technology investor — and a controlled experiment proves that even skilled practitioners get slower without operating model change — the debate is settled. The constraint is absorption. The solution is capability architecture.

On Our Radar

AI pricing inflection. Ignacio de Gregorio (TheWhiteBox) predicts AI subscriptions will reach $1,000+ per month as subsidies become unsustainable. Current pricing reflects investor subsidy, not true cost. Budget accordingly.

Apple's agentic strategy. WWDC will reveal Apple's four-layer approach: Siri as default agent, App Intents for developer ecosystem, MCP integration, and Gemini partnership for advanced reasoning. Platform sovereignty over model intelligence — the same principle we advocate for enterprises.

One Number

Fewer than 10% — the percentage of vertical AI use cases that ever make it past the pilot stage. Nine out of ten function-specific AI deployments die before reaching production. Not because the technology fails. Because the organisation cannot absorb the change required to make them work. (McKinsey/QuantumBlack, "Seizing the Agentic AI Advantage," June 2025)

The Transformer's Signal is published weekly by BusinessGPS. For the full analysis behind each signal, read The Capability Brief. For deep dives on specific topics, read The Way Pointer.