Card 07 · AI agents, explained
A routine: work that starts without anyone asking
A scheduled AI agent runs a saved instruction on a clock, in the background, and leaves the result where the team already looks. This is not a calendar scheduler: Monday's report exists before anyone remembers to ask for it.
01 — The ask
Work that waits to be asked slips.
The report is due Friday, and nobody types the request until Monday afternoon. A routine removes the typing, not the thinking: the instruction is written once and then followed.
02 — The trigger
A routine starts on a clock or an event.
Friday at two, or the moment the week's timesheet lands. Either way the start is a rule instead of a person's memory.
03 — The instruction
It is a saved prompt, not a person's habits.
The same words run every time, in the same order, with the same sources named. That is why two Fridays can be compared instead of argued about.
04 — The run
It runs where the agent already runs.
On the same account, with the same permissions, in the background. A routine does not get extra reach because nobody is watching it.
05 — The result
It leaves the result where the team looks.
A file in the Room, a draft in the shared inbox, a message in the thread the team already reads. Nobody learns a new app to find Friday's report.
06 — The failure
A skipped run has to be visible.
The useful question on Monday is whether Friday ran at all, so the run itself leaves a line. A quiet routine that stopped three weeks ago is worse than no routine.
07 — The difference
It is not a reminder. The work is done.
A calendar entry tells a person to start; a routine starts. The ten-person firm feels this as one less thing on a Friday list, not as another notification.
08 — Start here
Pick the thing everyone forgets on Friday.
Write the instruction as if you were briefing a new teammate, run it once by hand, then put it on the clock. The first week tells you whether the wording survived contact.
Every run leaves a line, including the Fridays that failed.
It runs at two with the week's timesheets, and the draft is waiting before anyone asks.
Tools for this
Every price and licence below is what the tool’s own page said on 18 November 2026, with two sources behind each record. There are no scores and no order of preference; where TalentOS is one of the tools it is marked ours and listed like the rest.
An automation platform whose MCP server exposes app actions to an AI client, running them on the app connections a team already has in Zapier.
For a team already paying for Zapier that wants its agent to act in the apps it has connected, and its recurring work on a schedule.
$19.99 per month (usage) · Closed source
An open-source scheduler for batch and data jobs, where each job is a Python file with a schedule, retries and a run history.
For a data or operations team that already runs Python pipelines and wants them on one schedule with one place to look when one fails.
$0 self-hosted (free) · Open source, Apache-2.0
GitHub's own runner platform, where a workflow on a cron schedule runs commands, tests or an agent without anyone starting it.
For a developer whose code already lives on GitHub and who wants scheduled or event-driven jobs reviewed like code.
$0 with 2,000 minutes a month (usage) · Closed source
GitHub Actions documentationGitHub plans and included minutes
An event-driven job platform for background work: step functions that retry, sleep, run on a schedule and keep their progress.
For a developer building AI work that takes minutes, where a failed step has to resume rather than start over.
$0 on the free tier (usage) · Open source, SSPL-1.0 with a future Apache 2.0 licence
A workflow tool where a saved workflow runs on a schedule or a trigger, with AI steps you can add to any workflow.
For an ops lead who wants recurring work automated without standing up a service or writing a scheduler.
20 EUR per month (usage) · Open source, Sustainable Use License
A durable execution engine: a workflow that survives crashes, restarts and deploys, and can keep running for days or wait for a person.
For an engineer who needs a long-running job that cannot silently lose its place halfway through.
$150 of free credits, then $50 per million actions · Open source, MIT