Data from the US Census Bureau on 583,000 businesses in 2018 shows AI adoption is slow: 8.9% use AI in any form, 2.8% use ML, and 2.5% use voice recognition
Will Knight / Wired : Tweets: @nxthompson , @quinnypig , @erikbryn , and @paul_scharre Tweets: @nxthompson : It may seem like AI is all the rage. In fact, only around 9% of US companies use any kind of AI, though that may soon change. As @erikbryn told @willknight, “there is a tidal wave in front of us.” https://www.wired.com/... HydroxyCoreyQuinn / @quinnypig : https://twitter.com/... https://twitter.com/... Erik Brynjolfsson / @erikbryn : Here's an article by @willknight in @WIRED describing a new paper that I co-authored with many others looking at the adoption of new technologies like #MachineLearning . We are still in very early days for most of these technologies. https://twitter.com/... Paul Scharre / @paul_scharre : “We are just in the very early days of adopting AI,” says Erik Brynjolfsson. “People should not think that the machine-learning revolution is petering out or is yesterday's news. There is a tidal wave in front of us.” https://www.wired.com/...
Context & Ripple Effects
When Will Knight published this Census Bureau analysis in August 2020, it landed as a corrective to the hype cycle: Erik Brynjolfsson co-authored the underlying paper and framed the moment to Wired as 'a tidal wave in front of us' — enormous potential sitting behind an adoption rate of under 9% across 583,000 firms. It became the baseline measurement everyone else has been measuring against since.
The arc since then validates both halves of that framing. By late 2022, McKinsey found 50% of businesses using AI, 2.5x the 2017 level — yet plateaued there for years. And the frontier has shifted from firms to workers: Gallup tracks daily individual use climbing quarter over quarter into 2025 while nearly half of employed adults still never touch AI.
First-order effects
- In 2018 the direct finding is concentration: with only 2.8% of businesses using machine learning and 2.5% using voice recognition, AI capability sits inside a small minority of firms, making early adopters the reference class for every vendor and consultant selling into the other 91%.
- For Brynjolfsson and the paper's authors, the dataset becomes the canonical citation for arguing that general-purpose technologies diffuse slowly even when their eventual impact is large — the 'tidal wave' claim now rests on measured baselines rather than anecdotes.
Second-order effects
- Vendors and cloud providers respond by productizing AI downward: the jump from 8.9% firm adoption to McKinsey's 50%-and-plateau reading suggests the easy gains came from packaged tools, while the stubborn half of the market needs integration work that off-the-shelf products don't solve.
- The plateau exposes a measurement and expectation gap — hype cycles priced in near-universal adoption, so buyers and investors recalibrate against survey evidence like Gallup's employee-level tracking, where workplace AI use doubled from 21% to 40% of employees between 2023 and 2025 but weekly-heavy use remains a minority behavior.
Third-order effects
- If firm-level adoption plateaus around half while worker-level usage keeps grinding upward, the structural shift is toward AI entering companies bottom-up through individual tool use rather than top-down through enterprise deployments — forcing IT, procurement, and governance regimes built for software licenses to absorb shadow adoption.
- The pattern fits Brynjolfsson's long-standing argument about general-purpose technologies: diffusion lags capability by years, and the productivity payoff arrives only after firms reorganize processes — which is why later analyses find roughly 40% of adults using AI while hard evidence of economy-wide productivity gains stays thin.
The trend: US AI adoption follows a two-track curve — rapid uptake by a leading half of firms and individuals, then a long, uneven diffusion tail — making the 2020 Census baseline the low-water mark against which every subsequent plateau and surge gets judged.