Routine

A routine is a named, repeatable agentic workflow that a team can invoke on demand. Routines are the building blocks of an institutional AI marketing practice — the equivalent of the 'playbook' in earlier-decade marketing operations, but with executable rather than purely descriptive content.

The practical anatomy of a routine has four parts: a name and clear-stated purpose that the operator can search for and select; an input contract that defines what the routine needs to run (the brief, the target, the constraints); a defined sequence of agentic steps (with handoffs, checkpoints, and review surfaces); and an output contract that defines what the routine produces and where it lands. Mature routine libraries also include version history, evaluation results, and recorded reasoning for the design choices behind the routine.

The routine framing is, in our view, one of the cleaner ways to give a marketing team durable institutional leverage from AI without requiring every operator to be a workflow engineer. The team's senior practitioners design routines; the broader team invokes them; the library grows over time; and the institutional knowledge accumulates in a place where it can be reviewed, improved, and audited. Teams that have built real routine libraries tend to find that their effective velocity climbs steadily without the corresponding rise in operator burden that less-structured programs produce.