Skip to content

AI Compass / Quick reads

Many agents in parallel – only if the task splits up

Automation Costs

More agents are not automatically better. They pay off when the work really splits into independent parts.

In depth: Long runs and parallel work – Claude Code's built-in tools

According to the docs, Claude Code can write a workflow: a script that controls several subagents at the same time, at most 16. It is triggered by the word ultracode in the prompt, or Claude plans one itself when /effort ultracode is on (docs checked). Jack Roberts draws the line: only for tasks that split into independent parts, not for debugging or work under five minutes (9 June). For facts, he has several skeptics attack a claim. For long runs, AI LABS adds: auto mode instead of switching off approvals, variants in their own worktrees, logs watched by the monitor (29 June). The cost is high; AI LABS cites about 1 million tokens for a small research task (3 June).

Tool and alternative

Everything is built in; on the Pro plan, workflows must first be switched on in /config. /workflows shows tokens per agent and can pause a run. According to the docs, agent teams are still experimental. The way it works has changed since June: in the videos it was 10 agents and a keyword, whereas ultracode is now a setting alongside the effort level.

Sources

Advertising in the source: Vanta (9 June), ScaleKit (29 June, 0:32–2:42), self-promotion for the paid community (3 June).

Want to try this with your team? That’s what the AI Lab is for. See the AI Lab