Former Anthropic researcher Jacob Coxon and economist Robin Hanson disagree on whether AI could rapidly automate much cognitive work, or whether its effects will take decades to spread through the economy. Their exchange also puts two very different approaches to AI regulation on the table.
Jacob Coxon Watch analysis
What happened
At a Y Combinator debate, Coxon argued that AI could automate 80% of cognitive labour within jobs by the end of 2028. Hanson challenged the forecast, pointing to past technologies whose economic effects took decades to unfold and to the organisational work that faster software does not automatically eliminate. BigGo Finance reports the exchange and its arguments.
Coxon proposed limiting the share of frontier companies’ computing power used to train more capable models, and requiring external safety-case reviews before training large models. Hanson favoured liability insurance as a way to make companies account for risk, while warning that regulators could become permanent and be captured. These are proposals from the debate, not adopted policies.
Why it matters
The disagreement is not simply about whether models will improve. It is about whether progress in AI capabilities translates quickly into economic change, and whether governments should act before serious failures or wait for clearer evidence. Those assumptions lead to very different ideas about what a sensible safeguard looks like.
Our read
The useful part is the collision between a dramatic timeline and a slower account of how economies absorb new technology. Coxon’s forecast is striking; Hanson’s reminder that organisations are more than their quickest individual tasks is a worthwhile check. Neither timeline settles the argument, but the policy proposals make the disagreement practical rather than merely speculative.
What to watch
- Whether evidence of AI-assisted research and work supports Coxon’s proposed timeline.
- How quickly organisations turn faster individual tasks into changes in overall output.
- Whether compute limits, safety-case reviews or insurance gain serious policy support.
Discussion spark: Should frontier AI companies face external limits before clear harms occur, or should regulation wait for stronger evidence that the technology is moving faster than the economy can absorb?
Sources and evidence
Independent WittyWires-operated tracker of Jacob Coxon. Not affiliated with or endorsed by Jacob Coxon, OpenAI or Anthropic.