The Bridge Ep. 18: Will AI Replace Jobs Or Create Them?

Summary

Ronnie Chatterji says OpenAI hired its first chief economist to study AI’s economy-wide effects across jobs, enterprise adoption, and the still-emerging AGI economy. He argues AI is already being adopted fastest by frontier firms and is spreading from coding into finance, legal, marketing, and people operations, but that the main effect so far is task reallocation and acceleration rather than mass displacement. The biggest labor-market concern is early-career workers, where AI may weaken the apprenticeship ladder before evidence becomes definitive. Chatterji also emphasizes that traditional GDP and productivity statistics miss much of AI’s value, including consumer surplus, new business models, and productivity gains from free tools. His broader thesis is that AI’s success should be judged by whether it drives growth, improves innovation in sectors like semiconductors and life sciences, and creates new categories of work rather than simply replacing existing jobs.

Notes

• OpenAI hired Ronnie Chatterji as its first chief economist to study AI’s economy-wide effects, especially jobs, enterprise adoption, and the still-undefined AGI economy.
• Chatterji says OpenAI wanted economics leadership that combined academia, public policy, and enterprise experience.
• His team is split into three pods: future of work, enterprise adoption, and the AGI economy.
• He frames OpenAI’s economics work as partly about questions standard tools cannot answer yet.
• The CHIPS and Science Act is presented as a new industrial-policy playbook for deep tech, including AI, quantum, synthetic biology, and energy.
• He says the US was making zero percent of advanced semiconductors before the CHIPS and Science Act.
• He says semiconductor buildout bottlenecks are workforce preparation and training for fabs and clean rooms.
• AI capex is described as having supported GDP over the last two years, with estimates ranging from roughly 0.25% to 1% of GDP.
• He argues AI is changing tasks inside jobs more than eliminating jobs immediately.
• Lower costs can increase demand for some services, so the labor effect may be reorganization rather than simple substitution.
• He says retraining and new onboarding will be needed as AI creates new professions and reorganizes existing ones.
• He wants coordination across government, business, schools, and community colleges for workforce transition.
• The biggest near-term labor risk, in his view, is for people starting their careers now.
• He worries AI could weaken the apprenticeship ladder for first-job entrants.
• He cites the Stanford paper Canaries in the Coal Mine, but says the evidence is not yet definitive.
• He still says STEM matters, especially formal logic and computer science.
• Human skills and humanities become more valuable because the key question shifts from how to do work to what should be done.
• He says leaders will still be needed because workers must learn how to delegate to AI.
• Financial advisors’ advantage may be the human relationship, not just information delivery.
• He expects healthcare and education to absorb more human demand as AI makes societies richer.
• Remote, rote, low-interaction jobs are the most exposed to full automation risk.
• He warns that AI gains may be concentrated by industry and geography, creating a K-shaped outcome.
• Targeted assistance and retraining are his preferred tools for a concentrated AI transition.
• He says OpenAI has released an industrial policy blueprint and that fiscal and monetary policy still matter.
• Measurement of AI in the economy is, for him, the most important policy issue.
• He says breadth of adoption matters less than depth of adoption.
• Enterprise surveys that only ask whether a firm uses AI miss intensity, use case, and ROI.
• He says O*NET is an incomplete way to classify modern work and AI exposure.
• Current labor classification systems may not match how work is actually performed now.
• AI-generated consumer surplus is large but mostly absent from GDP and productivity data.
• Free ChatGPT usage creates value that standard statistics fail to capture.
• He says AI may create new business models and other general-purpose effects that today’s statistics will miss.
• AI can be the first affordable expert for small businesses and people without access to consultants.
• He says AI can also help with health and family decision support, not just business tasks.
• Entrepreneurs are using AI mostly for marketing.
• AI helps solo founders add capabilities like marketing, finance, and web building without hiring immediately.
• He says the entrepreneurship trend may be correlated with AI but causality is not proven.
• Small businesses can turn AI-enabled formation into job creation as they expand.
• He says business impact should be measured by whether firms grow longer or scale faster with fewer employees.
• He expects incumbents, not just startups, to adopt AI aggressively to avoid being disrupted.
• Market selection matters: capital-intensive sectors are harder for small businesses, while local services leave more room to compete.
• He says agents are now part of the workforce.
• Frontier firms are using about 8x more tokens per user than median firms.
• The frontier-vs-median token gap has widened from 2x to 8x in five months.
• Coding was the first major agentic use case.
• OpenAI’s research and engineering teams saturated first, then finance, marketing, legal, and people operations caught up within about four months.
• He sees AI as more of an accelerant and complement than a substitute, at least for now.
• He says model evaluations have improved sharply, but the key variable now is enterprise adoption depth and realized returns.
• Enterprise AI is moving from brute-force token use toward disciplined task allocation across models.
• The key macro question is whether AI can drive economic growth.
• He says AI could boost growth by accelerating innovation in semiconductors, life sciences, and manufacturing.
• Two worries keep him up at night: capability overhang and whether AI gains are shared broadly.