Some jobs are too big for one Claude working start-to-finish. A workflow splits the work across a team of agents that run at the same time — one per file, per module, or per question — and then combines what they found. You describe the goal in plain language (this works in Claude Desktop too); the team divides and conquers. Think of it as going from one assistant to a whole room of them. This is an advanced, Power-Track-leaning capability — the natural-language ask works anywhere, but managing big runs (/workflows) and orchestration scripts is for heavier, technical jobs.
When one Claude works start-to-finish, you wait for each step in turn. A workflow changes the shape of the work: it splits the job across a team of agents that run at the same time, then gathers their results back into one answer.
The classic shape is fan-out, then gather — give each agent one slice (a file, a page, a question), let them work in parallel, and combine what they find. Some jobs add a verify pass, where a second wave of agents double-checks the first wave’s findings before anything is reported. You describe the goal in plain language; the team divides and conquers.
When a task is genuinely too big for one conversation to coordinate — a
codebase-wide audit, a large migration, a research question that needs
cross-checking — you can ask for a dynamic workflow. Claude writes an
orchestration script for your task and runs it across dozens to hundreds of
subagents in the background, verifying the results before it reports back. You
manage the runs from /workflows, and the heavy lifting happens while you do
something else.
It shines on work that is wide rather than deep: auditing a whole site, applying the same change across dozens of files, or asking one question of many documents. For a single tightly-coupled task, one focused Claude is still the right tool.
why it helps Turn a day of sequential grinding into minutes of parallel work.
examples- You start one by just describing the goal in plain language — the same in Claude Desktop or the terminal.
- Use it when work splits cleanly into independent pieces — one per file, page, or question.
- For a job too big for one conversation, ask for a workflow — Claude writes the orchestration script and runs it across many agents in the background (a Power-Track move).
- Ask for a single combined report at the end, so you read one summary instead of ten — many teams add a verify pass before reporting.
Questions people ask
- When should I use an agent team instead of one Claude?
- Reach for a team when the work is *wide* rather than deep — it splits cleanly into independent pieces, one per file, page, or question. For a single tightly-coupled task, one focused Claude is still the right tool.
- How do I start a workflow?
- Describe the goal in plain language — "review every page for broken links, in parallel, and give me one combined report" — which works in Claude Desktop or the terminal. For a job too big for one conversation, ask for a *dynamic workflow* and Claude writes the orchestration script and runs it across many agents in the background (a more advanced, Power-Track move).
- What's the difference between a workflow and a subagent?
- A subagent is one helper on one task. A workflow fans the work out across a *team* of agents running at the same time, then gathers their results into one answer. A workflow is many subagents working in parallel.
- Do I have to wait while a workflow runs?
- No. A dynamic workflow runs across dozens to hundreds of subagents in the background while you do something else; you manage the runs from `/workflows` and get one combined report when it's done.
- Is this a Desktop feature or a terminal one?
- Both, at different depths. The plain-language ask ("do this across every file, in parallel") works in Claude Desktop. Managing large runs and writing orchestration scripts is heavier, Power-Track work — useful when a job is genuinely huge.