Two years into the generative AI wave, the headlines have outpaced the data. Every vendor claims transformation; every board wants an AI strategy. But when you strip the narrative back to what the research actually shows, the picture is more nuanced — and more instructive — than the hype suggests.
Adoption Is Broad, But Value Capture Is Not
Generative AI adoption has gone mainstream. According to McKinsey's State of AI survey, the share of organizations using AI in at least one business function has roughly doubled in the last 18 months, and a clear majority now report using generative AI specifically. The surprise is on the value side: most organizations report no material EBIT impact from their gen AI efforts yet.
This is the central tension of 2025: adoption is easy, value capture is hard. The organizations that are pulling ahead aren't doing more pilots — they're doing fewer, deeper deployments tied to specific P&L outcomes.
The AI Index Shows a Widening Performance Gap
Stanford's AI Index Report 2025 documents the model-capability side of this story. Frontier model performance on benchmarks like MMLU, GPQA, and agentic evaluations has continued to improve rapidly, while inference costs have fallen by more than an order of magnitude year over year. That combination — better models at lower cost — is what's driving the gap between organizations that are operationalizing AI effectively and those still stuck in pilot mode. The tooling is no longer the bottleneck.
Where We Are on the Hype Cycle
Gartner's 2025 Hype Cycle for Artificial Intelligence places generative AI past the peak of inflated expectations and sliding toward the trough of disillusionment — exactly where leadership teams should expect to find it. What emerges from the trough are technologies with durable value: agentic AI, composite AI, AI engineering practices, and AI TRiSM (trust, risk, and security management). Expect budget conversations in 2026 to feel noticeably more disciplined than they did in 2024.
What Separates the Leaders
Deloitte's State of Generative AI in the Enterprise survey identifies a consistent pattern across high-performing adopters. Leaders invest disproportionately in data readiness, workforce enablement, and governance. They treat AI as an operating-model change, not a tool purchase. And they measure outcomes at the process level, not the pilot level.
Laggards, by contrast, tend to over-index on tool selection and under-invest in the change management that makes tools actually usable at scale. The result is the familiar pattern of "AI everywhere, outcomes nowhere."
The Employee View
BCG's AI at Work 2025 survey surfaces a useful counterpoint to the executive view. Employees who use AI regularly report meaningful productivity gains and, notably, higher job satisfaction. But the distribution is uneven: power users extract significant value, while the majority of employees use AI sporadically and see modest returns. Closing that gap — through training, workflow redesign, and targeted access to capable tools — is one of the highest-leverage investments companies can make heading into 2026.
What to Do About It
The playbook that emerges from the 2025 data isn't complicated. Concentrate investment on a small number of high-value use cases. Build measurement discipline so you can tell an adopted tool from a valuable one. Upgrade the data and governance foundations that let AI deployments scale beyond the pilot team. And treat workforce enablement as a first-class initiative, not an afterthought.
The organizations that enter 2026 with those fundamentals in place will compound advantages quickly. The ones that don't will keep running pilots.
Key Takeaways
- AI adoption is broad, but most organizations see no material EBIT impact yet
- Model capability is improving faster than enterprise operating models are adapting
- Gartner puts gen AI sliding toward the trough — expect more disciplined spending in 2026
- Leaders invest in data, governance, and workforce enablement — not just tools
- Employee power-users show the productivity gains are real when the workflow change is real