The AI productivity boom is confined to tech because developers have fewer adoption hurdles. Coding is a text-only medium with self-contained context in a codebase. In contrast, roles like marketing or law require complex data setup and workflow re-engineering, slowing down the productivity gains seen in macro-economic data.
Related Insights
Atlassian's founder suggests a model for AI's impact. In "input-constrained" fields like legal or support, AI drives efficiency on a finite set of tasks. In "creation-constrained" fields like software development, AI amplifies output on an infinite roadmap, leading to market expansion.
A dominant AI analytics company hasn't emerged because of user behavior, not technology. Analytics professionals have deeply ingrained workflows. Overcoming this inertia is a far greater adoption challenge than for simpler tasks like copy editing, slowing the entire category's disruption.
A new technology's adoption depends on its fit with a profession's core tasks. Spreadsheets were an immediate revolution for accountants but a minor tool for lawyers. Similarly, generative AI is transformative for coders and marketers but struggles to find a daily use case in many other professions.
Even if AI perfects software engineering, automating AI R&D will be limited by non-coding tasks, as AI companies aren't just software engineers. Furthermore, AI assistance might only be enough to maintain the current rate of progress as 'low-hanging fruit' disappears, rather than accelerate it.
AI's value is overestimated because experts view complex jobs as simple, solvable tasks. The real bottleneck is the unproductive effort required to build a custom training pipeline for every company-specific micro-task. Human workers are valuable precisely because they avoid this “schleppy training loop” by learning on the job, a capability current AI lacks.
Economist Tyler Cowen argues AI's productivity boost will be limited because half the US economy—government, nonprofits, higher education, parts of healthcare—is structurally inefficient and slow to adopt new tech. Gains in dynamic sectors are diluted by the sheer weight of these perpetually sluggish parts of the economy.
The slow adoption of AI isn't due to a natural 'diffusion lag' but is evidence that models still lack core competencies for broad economic value. If AI were as capable as skilled humans, it would integrate into businesses almost instantly.
Just as electricity's impact was muted until factory floors were redesigned, AI's productivity gains will be modest if we only use it to replace old tools (e.g., as a better Google). Significant economic impact will only occur when companies fundamentally restructure their operations and workflows to leverage AI's unique capabilities.
AI's "capability overhang" is massive. Models are already powerful enough for huge productivity gains, but enterprises will take 3-5 years to adopt them widely. The bottleneck is the immense difficulty of integrating AI into complex workflows that span dozens of legacy systems.
The focus on AI writing code is narrow, as coding represents only 10-20% of the total software development effort. The most significant productivity gains will come from AI automating other critical, time-consuming stages like testing, security, and deployment, fundamentally reshaping the entire lifecycle.
