What's this about?
Mollick: with agents we are at another large gap between AI abilities and public perception, and exponential gains mean the gap grows over time. Agents are now genuinely capable of long-running self-organized work, and that requires a new approach to AI.
Why would the gap widen rather than close?
Because perception updates linearly through experience and capability moves faster than experience accumulates. Someone who tried an agent six months ago has a mental model from six months ago, and the interval between their samples is longer than the interval between meaningful improvements. It compounds: the further behind your model is, the less likely you are to try again, because your outdated model says it is not worth it.
Why does this matter to you?
271 likes
- It is the argument underneath his essay on agency and agents, which asked for agents that reach out to humans more, not less.
- "Long-running self-organized work" is exactly what the Hugging Face incident demonstrated in the worst possible way.
- The practical implication for anyone evaluating tools: your last data point is probably stale, and it is staler than you think.