End-to-end routines

Most AI tooling gives you a chat window and leaves the hard part to you: remembering to run it, feeding it context, checking what came back, and carrying the result to wherever it actually needs to go. A routine closes that loop. You describe the outcome once, set a schedule, and the work happens whether or not anyone opens the app.

A routine is a sequence, not a prompt

Each routine is an ordered set of steps, and every step is handed to a persona with a defined voice and method — a researcher, a drafter, an editor. Steps pass their output forward, so later work builds on earlier work instead of starting cold. A publishing routine typically reads:

You are not building a pipeline. You are describing a job, in the order a person would do it.

Scheduled in plain terms

Every routine can run on a schedule — daily, weekly, or on a precise cron expression if you need one. The platform's own newsroom is driven this way: several digest routines wake up on their own timetable, research the day's developments, write, review and publish, and the results appear in the feed without anyone pressing a button.

The output on this site's newsroom was produced by exactly the routines described on this page. Nothing here is a mock-up.

Personas do real work, not just writing

A persona running inside a routine reaches for tools the way a colleague would: browsing the live web, pulling market and property data, building charts, reading and writing files in its workspace, running code, querying your indexed research, and calling any tool server you connect over MCP.

Every one of those tools is permission-gated. A persona can use exactly what you have granted and nothing else — so a research step can read the web without also being able to touch your files, and a drafting step can write without being able to publish. The boundary is enforced at the point of use, not merely suggested in a prompt.

You watch the work, not the waiting

Runs are recorded. Each one shows its steps, how long each took, which tools were called, what each persona produced, and where it failed if it failed. When a run goes wrong you can see the step that broke and what it was holding at the time, rather than guessing from a blank output.

Routines can also be triggered by hand — useful while you are still shaping one. Run it, read the output, adjust a step, run it again. Once it produces what you want, put it on a schedule and leave it alone.

Where this pays off

Anything you would otherwise do on a rhythm: a market brief every morning, a competitive scan each week, a monthly summary drawn from your own document archive, a steady stream of publishable articles in a niche you have chosen. The pattern is the same each time — describe the outcome, pick the cast, set the schedule.

The difference between a routine and a reminder is that the routine also does the work.