One request for Dots: make it the native default that a Dot spreads work across multiple agents working in parallel, instead of one agent doing everything one step at a time in a single thread.
DeepSeek just unveiled a Super Agent setup built on exactly this idea: the human sets the goal, multiple agents carry out the pieces at the same time, and the system keeps context in sync and brings results back for review. That is how Dots should work by default.
The win is speed and quality. Independent pieces run at once instead of waiting in line, and separate agents can check each other’s work instead of one thread grading its own homework. The user should not have to ask for parallel agents or manage the split themselves. Set the goal, let the Dot fan the work out, get one clean result back.
Michael Abamba
Small correction to my own post: it should not be the default for everything.
Dots should be smart enough to know when a job is big enough to be worth splitting up, and only then trigger it. And the Dot itself should not be the one spawning the agents. Work chats can already spawn subagents, so the flow should be: the Dot gives the instruction to the work chat, and the work chat spawns the subagents.
Each subagent owns a section. One builds this part, another builds that part, they coordinate with each other as they go, and the work happens in parallel instead of one thread doing everything in sequence. In theory that builds a lot faster, because the pieces are being built at the same time instead of waiting in line.
So less “every Dot runs many agents by default,” more “Dots recognize when parallel work would help and tell the work chat to fan it out.”
Michael Abamba