Dynamic workflows in Claude Code are editable JavaScript files that give full control over subagent order and logic. Even a pipeline of around 10 subagents is a valid use case, and they're more predictable than skills for repeatable tasks.
As part of Anthropic's expanded OneGov agreement with the US GSA, a four-week Utah pilot put Claude Code in front of 34 developers across 10 state agencies. 80% had a working environment within 30 minutes; 71% reported saving 20 or more hours over the pilot period.
With Max users in the US being cleared off the Claude Duet waitlist first, an Anthropic team member asked the community what they've been using it for.
Multi-account support for connectors like Gmail and Google Drive is on the roadmap after a user had to self-host an MCP server to work around the current single-account limit.
Functions like agent(), pipeline(), and parallel() are globals in Claude Code workflow scripts, so no imports are needed. You can ask Claude to generate a workflow file, save it from /workflows, and rerun it as a named slash command.
Hand-editing a dynamic workflow JS file is roughly equivalent to editing a skill's markdown, mostly tweaking values. The upside is token savings.
Users can already disable individual tools within a connector via the "Tool permissions" section, letting them allow email reading while blocking email sending, for example.
The /reload-plugins command lets users reload plugins mid-session without quitting the app, so work in one session can continue while a plugin is installed in another.
Bun's incentive structure has shifted now that it underpins Claude Code: where startup-era Bun optimized for growth, the current focus is production reliability, security, and stability.
If Claude keeps creating files or artifacts during chat, you can explicitly ask it to produce fewer, and that instruction can persist across an entire project as a standing preference.
