Pace of AI adoption
Experts like Tyler Cowen and Dario Amodei place optimistic (and realistic) GDP growth over the next few decades due to AI at 10-20% (and 3-5%), but this can only be realized if AI is widely adopted by the right markets. Large companies are pursuing AI adoption with insufficient aspirations. Companies like KPMG and McKinsey are trying to use AI to to accomplish the same objectives with less labor and resources, when they should be using AI to increase the pace of performance and enable growth and innovation. (There is a saturation effect in any market, which is especially true for management consulting, but we ignore that because we are nowhere near that limit.)
But there is too much resistance to AI adoption. When interviewing candidates, I ask how they would deploy or employ AI in their software processes. Their responses are very conservative and risk-averse — claiming they would use them for contained functions or designing interactions but not for the bulk of their work. I’m sure there’s a bias here as software candidates don’t know what their interviewers want to hear. But it’s telling that applicants in a field are weary of either employing revolutionary capabilities or admitting to do so.
In fairness, there’s a claim that the gains from AI-assisted coding are dubious or non-existent. In several studies (Git Clear 2025 and METR 2025, for instance), AI agents were found to have counter-productive effects on programming, though their human programmers paradoxically thought their efficiency had improved. This placebo feeling is worth exploring further. However, the ability of engineers at Anthropic, for instance, to generate a tool like Claude Code, about 90% of which was AI generated, speaks volumes to the capability. I think a balanced analogy would be comparing where we are now with a couch potato who has just started running and feels worse after their first workout than they did when they were on the couch. These are muscles and capabilities we’ve never used, and I would expect a learning curve.
Among the slow-moving industries (defense, healthcare, government) the risk of slow diffusion / adoption is greatest. Here, the potential gains are most impactful and will have outsized contributions to the GDP growth because these industries are ”addicted to labor.” Inconveniently, some companies and government groups will claim that safety-critical systems should be purposefully slow to adopt new technologies since we don’t yet fully appreciate the risks (cyber vulnerabilities, back doors, etc). But we should be reluctant to let these companies off the hook so easily.
There’s also a compounding effect of “any reason to refuse or denigrate AI is a good one.” This regime unites people who fear job loss, have security concerns (real or imagined), fear losing control of the systems for which they’re responsible, or are simply luddites stuck to the old ways. Worse, you don’t always know if these concerns are genuine or rationalizations from a deeper discomfort; and when you address one they may shift to another. Obviously, these issues are real and need to be adequately accounted for. But more often there’s hand-waving and an air of “we don’t know what cyber vulnerabilities [etc] are being included in codebases.” This laziness, in turn, is normally accepted from the non-adopters without an honest challenge.
I used to see this with some Christians who were scientists but, when they heard a fellow believer say that the human eye is so beautiful and complex it could only have been designed by a creator, would let it stand without any argument. There may be good reasons to believe in a divine being, but that’s not one of them. Similarly, we shouldn’t let luddites get away with baseless excuses for not implementing new technologies for any old reason. Claude and ChatGPT can even walk you through responsible ways to implement their coding suggestions and vulnerability assessments.
The companies and industries mentioned above need dedicated evangelists to constantly counteract the naysayers and help provide forward momentum when challenges arise.
