The last few months have made me rethink the size of the jobs I'm giving AI.
In July, Claude brought Cowork to web and mobile. Grok Bot arrived in August with its own computer. By September, Meta and OpenAI had introduced personal agents designed to keep working while you're doing something else. The models improved during the same stretch.
I've been spending a lot of time with AI agents. I'm interested in how much work they can finish before I need to step back in.
