This is the kind of change that slips past you, because it arrives as a patch note rather than as news. I want to write it down anyway, because the principle underneath it is bigger than the rule itself.
Today the team behind Claude Code changed a single word. The trigger word for dynamic workflows is no longer “workflow” but “ultracode”. You can still say “use a workflow for this”, but if you simply drop the word into a sentence, nothing fires up on its own anymore.
What that word set loose
The feature went out in research preview in late May. The idea: type “workflow” in a prompt, and Claude writes an orchestration script on the spot. That script then spins up a fleet of coordinated subagents working on your task in parallel. Not one assistant, but a whole crew that splits your job into pieces and runs them at once.
Powerful, for a genuinely complex job. The trouble sat in the trigger word itself. “Workflow” is an ordinary word. You use it in a hundred sentences where you want nothing of the sort. And that is exactly what happened: sessions quietly launching far too many parallel subagents and burning through their quota. It played a part in the limit resets Anthropic had to run in mid-May and early June.
Why explicit beats implicit
The fix is simple, and that is why it is good. A casual word should never set dozens of AI agents loose. By moving the trigger to “ultracode”, you now have to choose, on purpose, to summon that crew. Explicit over implicit.
It sounds like a detail, but it touches something I keep coming back to: with powerful tooling, it should be visible when you actually switch it on. A fleet of agents costs more tokens, more money, and gives you less grip on what is happening. That is not a reason to avoid it. It is a reason to do it deliberately.
The sober side
At the same time, this is still a research preview. The feature is young, the name changed within a week, and that will not be the last adjustment. Moving a trigger word does not fully resolve the deeper discomfort. Even summoning a fleet on purpose means you have to understand what you are setting in motion. The gain is not in the new word, but in the habit behind it: knowing when you put an army to work and when one assistant is enough.
That is the exact line I draw in my own work. AI may execute, as long as I see what it does and choose myself when it goes to full power.
More on this in my series on AI as an operator: how I left WordPress, the assistant that executes and what Claude can do now. Source: the announcement from @ClaudeDevs.