Daily Workflow

A working rhythm for leading a fleet of agents: starting the day, handing off work, and checking back in.

xSwarm agents work continuously and asynchronously — the pattern that works is closer to managing a team than pair programming. This is the rhythm that actually holds up in practice.

Start the day by reading, not writing

Before assigning anything new, check what’s waiting for a decision:

• Requests — anything you asked for that closed with evidence overnight, and anything still open.
• Blockers — questions an agent couldn’t answer alone. Each one carries options and a recommendation, so most take seconds to answer, not minutes.
• Activity — a live feed of what actually happened, not a summary someone wrote about it.

Answering what’s already queued is nearly always higher-leverage than starting something new — an agent stalled on a one-word answer is capacity sitting idle.

If you use a personal planning tool for your day — something that budgets time across development alongside everything else on your plate, not only code — the dev agenda belongs there as one input among others, not a separate planning ritual to maintain on its own. See Connecting Other Systems for how an external planner reads that agenda from xSwarm.

Handing off work

Two shapes, depending on how well-defined the work is:

• A request, when you know what you want and can state what would prove it’s done. This is the most durable way to hand something off — it survives a session ending, a context compaction, or being picked up by a different agent later, because it lives as a record, not a sentence in a conversation.
• A direct instruction to a running session, for anything more exploratory or where you want to stay in the loop as it develops.

Either way, state the outcome you want and how you’ll recognize it, not the implementation — an agent that’s told the destination can find a better route than one told each turn.

Checking on agents

Don’t interrupt to ask for status — it’s already visible. The live view shows what’s running right now; workers shows what’s registered on each machine; a project’s own page shows its blockers, its open work, and recent activity together. A session that’s gone quiet long enough shows as unknown rather than whatever it last claimed — see Coordination & Communication for exactly what each status does and doesn’t tell you.

What this replaces

Not a scrum standup, and not a PR review queue — there’s no ceremony that requires everyone present at once. The two structured records (requests and blockers) exist specifically so decisions don’t require synchronous time: you answer when you’re free, an agent proceeds when it has what it needs, and nothing sits waiting on a meeting that hasn’t happened yet.