Stanford's AI Index is the closest thing the industry has to an annual physical. The 2026 edition delivers one pairing that should reframe every AI conversation in your leadership team: organisational adoption has reached 88%, while the underlying global survey data shows just 7% of organisations report AI fully scaled. Adoption is no longer the differentiator. Scale is.
What You Need to Know
- 88% of organisations use AI somewhere, up from 78% a year earlier. Adoption, as a metric, is saturated and no longer tells you anything about competitive position.
- Just 7% report AI fully scaled, and only about a third have begun scaling at all. The distance between "we use AI" and "AI runs part of our business" is where nearly everyone is stuck.
- Agent deployment is still in the single digits across most business functions, despite a year of agent headlines.
- Documented AI incidents rose 55% in a year, to 362 in 2025. Deployment is accelerating faster than the discipline around it. That divergence is the strategic risk.
Augmentation Is the Comfortable Plateau
The Index describes the current enterprise phase precisely: AI layered onto existing processes as augmentation. It drafts the document, summarises the incident, suggests the test. The human workflow around it is unchanged.
There's nothing wrong with augmentation. It's real value and the right first step. The problem is that it's also a plateau with a view, and most organisations have stopped climbing. Redesigning a workflow around AI (changing who does what, what gets automated, where judgement concentrates) is harder than turning on a copilot, which is exactly why only 7% report being fully scaled. The 88% is the crowd. The 7% is the competition.
We called this the enterprise AI execution gap back in April, and this year's numbers put a figure on each side of it.
The Incident Curve Is the One to Act On
Alongside the adoption charts sits a harder number: documented AI incidents (systems causing harm, failing in production, or producing unintended outcomes) rose to 362 in 2025, up from 233 the year before. A 55% increase in a single year, during exactly the period when deployment accelerated.
Read the pairing honestly: usage is compounding faster than the discipline around it. Agent deployment sitting in the single digits per function isn't just caution, it's a market that has learned augmentation is safe and autonomy needs engineering it hasn't built yet.
The incident curve is also the risk most within your control. Most production AI failures we're brought in to diagnose trace to grounding: models answering from general knowledge when they should be answering from the organisation's data, with citations, inside guardrails. Retrieval, grounding, and verification are engineering disciplines, not model roulette. That's a solvable problem with a known playbook, which is more than can be said for most items on a risk register.
Closing the Gap Is a Choice, Not a Purchase
The scale gap doesn't close by buying more licences. In our delivery work, the organisations that cross it share three habits:
- They pick functions, not features. One process, redesigned end to end with AI inside it, measured against a baseline. Then the next.
- They put trust infrastructure in early. Grounded answers with sources, human checkpoints where stakes are high, and monitoring that catches drift. The incident curve shows discipline lagging deployment; the fix is building systems whose outputs can be checked, not asking people to have faith.
- They treat adoption as a people programme. Tools scale instantly; behaviour doesn't. The augmentation plateau is partly a change-management artefact, and no amount of platform spend fixes what is fundamentally a people problem.
The 2026 Index reads like a market waiting for its second act. Everyone's on stage. Almost nobody is performing at scale. For NZ and Australian firms, that's unusually good news: the global frontier is closer than it looks, and it's reachable with discipline rather than budget.