Agent SwarmPattern: Per-Customer Working Directories
Give each top account a persistent folder in agent-fs where agents accumulate context across months. A 6-month-old note like "they care about EU data residency" is what makes the next report feel personal.
Give each top account a persistent folder in the shared filesystem (agent-fs) where agents accumulate context across months — not just within a single session.
What it is
A durable directory per account (e.g. customers/<account>/) in agent-fs, holding running notes, integration history, open questions, recent activity, and meeting transcripts. Any agent in any future session can read it to answer "what's the state of this account?" without re-deriving everything.
Where we use it
- Proactive customer support — each top account has a directory. A post-call Granola transcript appends automatically; the next report-generation task reads months of accumulated context.
Why it works
- Agent memory is good for swarm-wide learnings; a working directory is better for account-specific, document-shaped context (notes, transcripts, history) that you want to read and edit as files.
- It turns "what changed since their last touchpoint?" from an impossible question into a diff against the directory.
- It survives session resets, model changes, and agent hand-offs — the context lives in files, not in one agent's head.
How to apply
- One directory per account, predictable path. Keep a short
README/indexat the root summarizing current state. - Append, don't overwrite — transcripts and notes are an audit trail.
- Separate internal notes from customer-facing drafts so a redaction pass is easy.
- Pair with the HITL gate: the directory feeds the draft; a human approves the send.
Used in
Pattern: HITL Gates (Human-in-the-Loop Approval)
Pause a workflow mid-flight until a human approves in Slack. The agent does the work; the human owns the irreversible decision (sending, merging, deleting, posting publicly).
Pattern: No-op When Nothing Changed
A scheduled workflow that first asks "did anything actually happen?" and skips silently if not — so it never produces empty reports, noise commits, or training-data-style "nothing to report" messages.