Scheduled AI Agents: Automate Recurring Work

Some work doesn't need a person to kick it off — it needs a clock. Scheduled AI agents run your recurring tasks automatically, on the cadence you set, without anyone remembering to hit "start."

Why scheduling matters for AI agents

Most AI agent demos show a person typing a prompt and waiting for a response. That's fine for one-off questions. It breaks down for the work that actually eats up ops, marketing, and finance teams: the Monday report that has to go out every Monday, the invoices that arrive every night, the research summary that's due every Friday whether or not anyone remembers to ask for it.

If an agent only runs when a human remembers to trigger it, it's not automation — it's a shortcut that still depends on someone's memory. Scheduling closes that gap. Once a pipeline is built and approved, it keeps running on its own, and the humans on your team only get involved where governance says they should.

This matters more as you move from single agents to multi-agent pipelines: a research → draft → review → publish chain is far more useful running unattended every week than triggered by hand each time. Scheduling is what turns a pipeline from a tool you operate into a process that operates itself.

Cadences and triggers: two ways to start a run

AgentWorks supports two ways to kick off a scheduled agent or pipeline, on Pro and above:

• Time-based cadence — daily, weekly, monthly, or a custom schedule you define. Set it once and the run happens at the same time going forward.
• Inbound webhook trigger — a run starts when something else happens: a form submission, a CRM field update, a CI pipeline step completing. No fixed clock, just an event.

Both approaches use the same underlying agents and pipelines you'd otherwise run manually from the chat workspace or build from AI agent templates. Nothing changes about how the agent behaves — only how it gets started. A webhook-triggered run goes through the same steps, the same approvals, and the same cost tracking as a manually started one.

Combining both is common: a weekly cadence for the routine report, plus a webhook trigger for the exception case — a new deal closing in your CRM, say, that should fire off a different pipeline immediately rather than waiting for the next scheduled slot.

Example recurring workflows

A few patterns show what scheduling is actually good for:

• Weekly research-to-draft pipeline. Every Friday, a pipeline gathers research on a topic, drafts a post from it, routes it to a reviewer, and holds for approval before publishing. The person who used to spend an hour each week doing this now spends a few minutes reviewing.
• Nightly invoice processing. Each night, an agent picks up new invoices, extracts the relevant fields, checks them against expected values, and flags anything that doesn't match — ready for finance to review first thing in the morning instead of starting from a blank inbox.
• Monday-morning reporting. A pipeline pulls the past week's numbers from your systems, assembles a report, and delivers it before your Monday standup, so the meeting starts with numbers already in hand instead of someone scrambling to pull them together.

None of these require a person to remember to start anything. They require the pipeline to be built once, connected to the right tools — Slack, Gmail, HubSpot, Pipedrive, Google Workspace, or a custom system via MCP, REST API, or webhooks (see integrations) — and scheduled.

Governance doesn't pause for scheduled runs

A run that happens automatically, unattended, at 2am is exactly the kind of run you want the most oversight on, not the least. Scheduling doesn't switch any of that off.

Every scheduled run goes through the same checks as a manual one:

• Each step gets classified for EU AI Act risk before it executes.
• PII is redacted at the gateway, not left to the agent's discretion.
• Human-in-the-loop approval still applies where your governance rules require it — a state-changing step (sending an email, updating a record, publishing content) still pauses and waits for a person, schedule or no schedule.
• Every step is written to an immutable audit trail, so a run that happened while nobody was watching is still fully reviewable afterward.

This is the practical difference between "automated" and "unsupervised." Scheduled runs are automated — they start themselves — but they're not unsupervised, because the same EU AI Act readiness controls apply whether a human or a clock started the pipeline. If a step in your nightly invoice run would normally need sign-off, it still needs sign-off; it just waits in your queue until morning instead of interrupting someone at night.

Cost transparency on every scheduled run

Scheduling doesn't create a black box around spend. Every scheduled run debits your transparent € wallet exactly like a manual one, with full per-step token transparency — you can see what each step in the pipeline cost, not just a total for the run.

Tokens are billed at cost plus 10%, and the AUTO router picks the most cost-effective model for each step, so a nightly run doesn't default to the most expensive option out of habit. Because scheduled runs happen without someone watching in real time, this visibility matters more, not less — you should be able to open the wallet the next morning and see exactly what last night's run cost, step by step, the same way you'd check transparent pricing before committing to a plan.

Getting started with scheduled agents

Scheduling is available on Pro (€39/mo, €10/mo balance included) and above, including Team (€49/seat/mo, with added collaboration features) and Enterprise (custom). It is not available on the Free plan (€0, €5 one-time credit, access to 50+ agents) — Free is for trying agents manually before you commit a recurring workflow to a schedule.

To set one up: build or select the agent or pipeline you want to run, decide whether it fires on a cadence, a webhook, or both, and confirm which steps need human approval before they execute. From there, it runs on its own — you just check the wallet and the audit trail to see what happened.

Summary: Scheduled AI agents and pipelines run on a cadence or a webhook trigger, on Pro and above, with the same EU AI Act governance, PII redaction, human approval, and per-step cost transparency as any manually run agent — so automating the work doesn't mean losing sight of it.

Frequently asked questions

What's the difference between a scheduled agent and a webhook-triggered agent?

A scheduled agent runs at fixed times you define — daily, weekly, monthly, or a custom cadence. A webhook-triggered agent runs whenever an external event fires, such as a form submission, a CRM update, or a CI step completing. You can use either or both for the same pipeline, depending on whether the work is time-based or event-based.

Do scheduled runs skip human approval steps?

No. Scheduled and webhook-triggered runs go through the same governance as manually started runs. State-changing steps still pause for human-in-the-loop approval where your rules require it, every step is risk-classified under the EU AI Act, and PII is still redacted at the gateway. The run being unattended doesn't remove the checks.

Which plans include scheduling, and what does it cost to run a scheduled pipeline?

Scheduling is included on Pro (€39/mo) and above, including Team and Enterprise; it's not available on the Free plan. Each scheduled run is billed like any other: tokens at cost plus 10%, debited from your € wallet, with per-step transparency so you can see exactly what a nightly or weekly run cost.