AI you ask becomes AI that brings you things.
A Scheduled Agent wakes on your cadence — every weekday at 8, Mondays at 9, the first of the month — grounds itself in your workspace memory, does a defined job, and leaves a durable result. Every single time.
The alarm fires
On its cron schedule, the agent wakes — reserving its run budget before it does anything.
Loads your memory
It grounds in the workspace — profiles, indexes, the target artifact, and recent run history.
Runs its task
It pulls live data, reasons, and produces the output it was configured for — under its caps.
Durable output
A new artifact version or a thread — announced in your channel, with a full trace.
Four ways an agent wakes up.
Boardbox agents share the same memory, governance, and budgets. What differs is the trigger. Scheduled is the recurring worker — the one that runs on a clock, whether you’re watching or not.
Channel
Scheduled
Reactive
Deep
Scheduled and Reactive are close cousins — both autonomous, both governed, both leave a run record. The difference is what wakes them: Scheduled runs on an alarm clock, asking “anything change since yesterday?” Reactive runs on a nervous system, told “this just changed” the instant it does. Scheduled is your operating cadence; Reactive is your reflexes.
Bring me a document, or start the conversation.
Every Scheduled Agent does one of two things with its run. The mode is the difference between a living artifact and a recurring thread.
It brings you a document.
The agent produces or updates a living artifact. Each run is a new version— so the artifact’s own history becomes the agent’s long-term memory of every prior run.
Daily ad performance — Jun 2
It starts the conversation.
The agent posts into a channel and opens a thread with a visible headline. Best when your reply matters — check-ins, journaling, standups, retros, recurring questions.
It wakes up inside a workspace, not a blank prompt.
A normal scheduled reminder fires a static message. A Scheduled Agent grounds every run in durable memory — so it knows what changed since last time.
Grounded in your memory.
Before it does the job, the agent loads what it needs: your personal profile, the workspace and channel index, the target artifact and its recent versions, pending proposals, and — for governed monitors — the plan’s watch conditions.
However the idea arrives, it becomes a job.
You always review the proposed setup — name, schedule, mode, persona, tools, channel, and budget — before the agent exists.
Every run has a budget. Every run leaves a trace.
You pick a tier, not a token count. Runs reserve budget before executing and settle to actual cost after — and if quota’s short, the run is skipped and surfaced, never failed in silence.
It shows its work before it earns autonomy.
The deepest agents run under an explicit autonomy policy: what auto-applies, what must escalate, daily action ceilings, spend caps, and an expiry. It doesn’t ask you to trust it — it shows what it would do, and how recent history would have behaved under the policy.
AI on a timer, inside an operating system.
The moat isn’t “AI runs on a schedule.” It’s that the schedule fires inside a workspace with durable memory, channel placement, real output, and approval boundaries.
The standing jobs that run your week.
Put your week on autopilot.
Describe a Scheduled Agent in plain language, pick its cadence, and let it bring you what matters — grounded, governed, and on time.