June 19, 2026
First it tanks, then it thanks
The Productivity J-Curve [pdf] (2018)
Why shiny new tech can look useless at first — and the comments got spicy
TLDR: The paper says breakthrough tech like AI can seem disappointing at first because companies must do lots of hidden setup work before the gains show up. Commenters then turned it into a brawl over whether that still fits today’s AI rush, plus a side quest about people confusing “GPT” with ChatGPT.
A 2018 economics paper is suddenly back in the spotlight for one deliciously counterintuitive claim: big breakthrough tech can make productivity look worse before it looks better. The authors argue that when businesses adopt game-changing tools like artificial intelligence, they first sink time and money into invisible stuff — retraining workers, redesigning workflows, inventing new products — and those gains don’t show up neatly in the stats right away. In plain English: companies may be doing a ton of work behind the scenes before the payoff finally arrives.
But the real fireworks were in the comments, where readers split into camps almost immediately. One crowd jumped in with the classic “actually, this doesn’t fit today’s AI boom” argument, saying the paper’s assumptions fall apart when everyone is fighting over limited chips, storage, and electricity. Another commenter practically grabbed the mic to do emergency terminology control, warning people that “GPT” here means “general purpose technology,” not ChatGPT, and that “productivity” means economic output, not whether you finally cleared your inbox. Yes, the comments section became Economics Class With Heckling.
And then came the comic relief: one drive-by critic declared the paper’s real lesson was never typeset research in MS Word, which is the kind of petty scholarly shade the internet lives for. Others tried to connect the paper to fresh AI productivity chatter, arguing the long-promised payoff may finally be arriving. So is this sober economic theory, a delayed AI victory lap, or just another chart people are using to defend their priors? The community, naturally, chose all three.
Key Points
- •The paper argues that general purpose technologies such as AI require large complementary intangible investments, including business process redesign, new business models, and human capital development.
- •These intangible investments are often poorly measured in national accounts, which can cause early underestimation of output and productivity.
- •The authors develop a model that produces a "Productivity J-Curve," where measured productivity initially falls or is understated before later rising as benefits are realized.
- •The paper applies the model to AI-related intangible capital and its effects on measured total factor productivity and output.
- •A historical analysis in the paper finds substantial and ongoing effects from software-related intangibles and smaller effects from computer hardware-related intangibles.