Bart Mol(@Bart_Mol):Just learned that this is called a dynamic workflow. I really enjoy the way it works. Your main agent stays responsive and uses way less context that normal. I instructed it to also use Sonnet and Haiku subagents. It worked for quite a while (hours) and the result was better than when a single agent with some subagents worked for a prolonged period of time. Less drift. I tried this before with Fable and that just used too much tokens. With Opus I used around 20% of my weekly budget on max20 for a 4 hour session.

2026.10.09 07:29

Just learned that this is called a dynamic workflow. I really enjoy the way it works. Your main agent stays responsive and uses way less context that normal. I instructed it to also use Sonnet and Haiku subagents. It worked for quite a while (hours) and the result was better than when a single agent with some subagents worked for a prolonged period of time. Less drift. I tried this before with Fable and that just used too much tokens. With Opus I used around 20% of my weekly budget on max20 for a 4 hour session.

Bart Mol@Bart_Mol

2026.10.08 10:07

I really like this overview and structure Claude Code creates when using ultracode. It's way better to get an idea what's happening. I have the feeling the subagents also don't use that much tokens because their task is well defined.