Claude Code Switches Trigger Word From "workflow" to "ultracode"
Anthropic's official "Claude Developers" account (@ClaudeDevs) announced on June 3, 2026 that it has changed the trigger word for Claude Code's Dynamic Workflows from "workflow" to "ultracode." The change aims to prevent the common term "workflow" from accidentally launching dozens to hundreds of parallel agents in everyday prompts.Source
Users can still kick off a Dynamic Workflow by explicitly saying "use a workflow for this," but when the context makes clear that "workflow" is being used in another sense (for example, a general business process), Claude will no longer start a workflow automatically. To trigger one explicitly, users are encouraged to say "ultracode." Anthropic said the change was based on user feedback and expressed its appreciation.Source
The feature affected by this adjustment, Dynamic Workflows, was released as a research preview around May 28, 2026. It lets Claude generate its own orchestration scripts and spin up tens to hundreds of parallel subagents to handle large-scale tasks.Source It is designed for software engineering tasks too large for a single agent, such as major refactoring, full codebase audits, and migrations spanning multiple files. Unlike conventional fixed multi-agent setups, it is characterized by Claude dynamically generating, executing, and verifying workflows according to the task at hand.Source
Because the word "workflow" itself served as the trigger, this frequently used term in code and everyday conversation could unintentionally launch large numbers of parallel agents, leading to token consumption and session limits. One shared case described 130 Opus 4.8 agents launching after someone typed "workflow" late at night.Source The change reduces accidental launches while letting intentional use be triggered clearly via "ultracode" or an explicit instruction.Source
The feature is available in Claude Code v2.1.154 and later, on Max/Team/Enterprise plans (Pro plans enable it via /config), as well as the Claude API, Amazon Bedrock, Google Vertex AI, and Microsoft Foundry.Source It can be triggered explicitly by including "workflow" or "ultracode" in a prompt, or automatically by specifying /effort ultracode, which combines the highest level of reasoning depth (xhigh reasoning effort) with automatic workflow orchestration. A single request can chain multiple workflows as an understand-modify-verify loop, with subagent counts referenced at 130 in cases and capped at 1,000.Source
Compared with rivals, while many agentic coding tools including OpenAI's Codex adopt fixed multi-agent configurations, Anthropic differentiates itself by having Claude write its own workflows, drawing particular attention in combination with the Opus 4.8 model.Source On Medium and elsewhere, explanations of agent swarms have appeared one after another, including a case of refactoring 750,000 lines of code in 6 to 11 days.Source
The official post drew many reactions in a short time, mostly positive. Common praise included "workflow was too common a word for developers," "ultracode is better because it's more intentional," and "the risk of accidentally launching hundreds of agents has been reduced." Meanwhile, some voiced requests such as "the name ultracode is a bit awkward" and "please allow custom trigger words in the future." Use cases discussed include large codebase migration, multi-file refactoring, and deep research, and even non-engineers called it "an important switch for seriously delegating work." Still, as it remains a research preview, some note that high token consumption and session limits still require caution.