Artificial Intelligence and Economic Growth

The economic modeling in this NBER paper is intimidatingly rigorous, but the core premise is fascinating for anyone building software. Aghion, Jones, and Jones ask what happens to long-run economic growth when AI automates not just physical labor or basic knowledge tasks, but the actual process of scientific discovery and invention itself. As a technical founder, this strikes close to home. We are already using LLMs to write code, generate tests, and refactor architecture. If we can automate the bottleneck of R&D—the human constraint of generating and validating ideas—growth models could shift from linear to exponential.

However, the authors introduce a crucial sanity check: Baumol’s cost disease. Even if we automate 99% of our processes, the rate of economic growth will ultimately be constrained by the remaining 1% that cannot be automated. This is a profound lesson for running a startup. We can optimize our codebase and automate our pipelines as much as we want, but the ultimate rate-limiting step of our company is still human alignment, fundraising, and deeply understanding our users. Technology might accelerate the execution, but the bottleneck remains human-centric.

What stuck with me

• Automating the automation: The ultimate economic accelerator is using AI to automate the process of invention and research itself.
• Baumol's shadow: Overall economic growth will still be constrained by the slowest, hardest-to-automate sectors of human activity.
• Capital accumulation shifts: Rapid automation will likely increase the economic share of income going to capital owners over labor.