I think that the biggest thing is that it's useful for search engine replacement if you can be bothered to actually check the links it cites.
For coding it's getting good, but not as good as people say it is. It's mostly much more scalable due to parallelization being cheaper and easier to attain. I.e. if you run 19 agents in parallel with one human vs train 20 humans, the agents will always win on consistency. But the skills of the single human are infinitely more impactful in the agentic setting than the skills of all humans combined in the human team setting: there are no complementary skills in the former setting, no real synergy; just scale and cheap restarts.
I think that the biggest thing is that it's useful for search engine replacement if you can be bothered to actually check the links it cites.
For coding it's getting good, but not as good as people say it is. It's mostly much more scalable due to parallelization being cheaper and easier to attain. I.e. if you run 19 agents in parallel with one human vs train 20 humans, the agents will always win on consistency. But the skills of the single human are infinitely more impactful in the agentic setting than the skills of all humans combined in the human team setting: there are no complementary skills in the former setting, no real synergy; just scale and cheap restarts.