Reading the Stanford AI Index 2026 Against the Canary Papers
Reference: The Stanford 2026 AI Index Report (Download PDF)
Stanford's Institute for Human-Centered AI released the ninth edition of the AI Index last week. It is 423 pages of independent, source-labeled data on where AI stands in 2026: capability, adoption, economics, workforce, public sentiment. I read it against the first two Canary Papers to see where the data agrees with the thesis, where it sharpens it, and where it does not.
The short answer: the signals hold. In several places, they get louder.
Capability is accelerating, not plateauing
Paper 01 argued that we have crossed a threshold where capability improvements are compounding faster than institutions can absorb them. The AI Index opens with essentially the same claim in its first top takeaway: "AI capability is not plateauing. It is accelerating."
The numbers behind that sentence are sharper than anything in the paper. On SWE-bench Verified, a coding benchmark, frontier model performance rose from 60% to near 100% of the human baseline in a single year. Humanity's Last Exam, built to be difficult for AI and favorable to human experts, saw 30 percentage points of improvement in twelve months. Global compute capacity has grown 3.3x annually since 2022. The report's own framing: "Evaluations intended to be challenging for years are saturated in months."
None of that proves Hassabis right about a century in a decade. But it puts the burden of proof on skeptics in a way it did not carry a year ago.
Jagged intelligence, now in the headline data
Paper 01 described AI as jagged: strongly capable in some domains and strangely weak in others. The AI Index uses the identical phrase and dedicates a top-line takeaway to it.
The concrete example is better than the ones I used. Gemini Deep Think won a gold medal at the 2025 International Mathematical Olympiad. The same class of models reads analog clocks correctly 50.1% of the time. Humans manage 90.1%. Robots succeed at 89.4% of manipulation tasks in simulation and 12% of household tasks in reality.
The jaggedness is real, it is measurable, and it is exactly the pattern the paper predicted would fuel skepticism even as the frontier advances.
The diffusion gap is now the story Stanford tells too
Paper 02's central claim was that the gap between what AI can do and what organizations are actually doing with it is the defining feature of this moment. The AI Index devotes an entire highlight section to "Measuring Signals of AI Diffusion." The framing is the same.
The numbers line up:
88% of organizations now use AI in at least one business function
AI agent deployment remains in the single digits across nearly all functions
Generative AI hit 53% population adoption in three years, faster than the personal computer or the internet
The United States, despite leading in investment and model development, ranks 24th globally in population-level adoption at 28.3%
Access is widespread. Integration is rare. The constraints, as Paper 02 argued, do not apply evenly across geographies, sectors, or organizations.
The front door is narrowing, and the data is stronger than I had
Paper 02 reported that hiring of younger workers aged 22 to 25 had slowed by roughly 14% in AI-exposed occupations since ChatGPT launched. The AI Index goes further. Drawing on Brynjolfsson et al. 2025, it shows that employment for software developers aged 22 to 25 has fallen nearly 20% from its 2022 peak, while headcount for older developers continues to grow. The chart shows the divergence opening cleanly in early 2023 and widening through 2025.
The structure is still standing. The doorways are getting smaller. The data has caught up with the language.
A note on the title
The Brynjolfsson paper that produced the 20% figure is titled "Canaries in the Coal Mine: Six Facts about the Recent Employment Effects of AI." The resonance with this series was unintentional but worth acknowledging. Two independent lines of inquiry, using different methods, reached for the same metaphor to describe the same phenomenon.
That is itself a signal.
Where the Index adds something the papers did not
Two additions worth flagging for the rest of the series.
The expert-public gap. When asked whether AI will positively affect how people do their jobs, 73% of experts say yes. Only 23% of the public agrees. That is a 50 percentage point gap, larger than most partisan divides in the same survey. Similar splits appear on the economy (69% vs 21%) and medical care (84% vs 44%). Paper 01 argued that the future negotiates. This is what the negotiation looks like in numbers.
The J-curve. US productivity growth reached 2.7% in 2025, nearly double the 1.4% decade average. But a parallel survey of 6,000 executives found widespread adoption with minimal realized productivity gains to date. Brynjolfsson frames this as a J-curve: organizations absorb the cost of integration before the returns appear. It is the clearest macro signal yet that integration, not access, is where the value lives.
The question that remains
Paper 02 closed with a line that continues to stay top of mind: "The rate of diffusion has changed. The question is whether the rate of rethinking can change with it."
The AI Index 2026 has given us a year of independent evidence. The rate of diffusion has indeed changed. Capability is outrunning measurement. Access is outrunning integration. Public trust is running behind both.
The rate of rethinking is the variable leadership controls. Stanford's data suggests most organizations are not yet moving it.
That is the signal. And it is getting harder to miss.
The Canary Papers
The Canary Papers is a six-essay series drawn from executive conversations on AI adoption and strategic change. Named after the early detection systems once used in coal mines, the series focuses on signals rather than headlines, where capabilities are compounding, where organizations are lagging, and where competitive gaps are quietly widening. Each paper examines a distinct pressure point, from displacement timelines to diffusion barriers and trust costs. The aim is not prediction, but disciplined clarity: to help leaders recognize structural shifts early enough to act with intention rather than react under pressure.
