What’s interesting about AI agents is the unit of a task is just going to keep on growing in size over time. In code, in the past 2 years, we’ve gone from autocompleting a couple lines of code in a second, to writing hundreds or thousands of lines of code in minutes, to writing tens of thousands of lines of code over hours. The same trend is happening right now in most other fields of knowledge work. Outside the space some may wonder “why isn’t AI getting cheaper or faster?” Yet, everyone in the industry knows that we’re now able to throw more compute at a problem to solve the next harder one. As soon as we think we’ve hit an optimal point of quality, price, and speed, the use case potential just expands as we get more model breakthroughs. The last problem we could solve looks trivial by comparison, and we’re onto the next complex task. This dynamic will likely keep happening for the foreseeable future given we’re nowhere near the plateau of capability, so the task units will just keep growing and growing.