Skip to main content
The Human Bit
Ask Bit
← All playbooks

Practical playbook · comfortable · 20 minutes

Organise recurring work

Keep the repeated context, inputs and review points in one place so a routine gets easier without quietly running itself.

This is for you when
You come back to the same report, project, client or review and set the whole thing up again from scratch every time.
What you will leave with
A reusable brief, a source set that stays current, and a clear human checkpoint on every cycle.

The Human Partnership

Use the capability. Keep the part that needs a person.

Recurring AI work should reduce repeated setup, not remove ownership. The real design task is deciding what context persists, what can run automatically and where a person must intervene.

AI contribution

Choose a capability for a real job.

Each capability removes a different kind of effort. Its value depends on the human context beside it.

01 · capability

Projects and persistent workspaces

Projects can keep approved instructions, files and related conversations together so a weekly report, client review or project update does not restart from a blank page each cycle.

The person contributes
The process owner maintains the source set, separates clients or matters, removes obsolete instructions and decides what context is safe and useful to retain.
Do not assume
Stored context can become stale, leak across poorly separated work or preserve a past decision long after the operating reality has changed.
02 · capability

Scheduled tasks, triggers and monitors

Scheduled features can prepare briefings, monitor a topic, check for changes or produce a recurring first draft at an agreed cadence without someone manually starting every run.

The person contributes
The owner chooses cadence, trigger, approval mode, escalation rule and stop condition, then reviews whether the automation still creates value.
Do not assume
Scheduled products differ in access to files, connectors and fast moving data. Some prepare responses early, some cannot reach project files, and none should be assumed to run with perfect continuity.
03 · capability

Connected tools and action workflows

Connectors can read approved email, calendar, drive, chat or project systems and may create or update records, allowing the routine to begin from live operational information.

The person contributes
A person grants the smallest useful permission, defines which actions require approval and remains responsible for messages, records or system changes made externally.
Do not assume
Inherited account permission can expose far more than the task needs, and a technically successful action can still be inappropriate, mistimed or based on stale context.

Human expertise

The resources already inside the work.

Experience, relationships, definitions and accountability are inputs to the workflow, not a final check after AI has finished.

Process ownership

You know why the routine exists, who depends on it, which input is authoritative and what should happen when the normal process breaks.

Exception judgement

You can recognise unusual cases, changed definitions, sensitive events and moments when the standard workflow should pause rather than continue confidently.

Operational accountability

You decide permissions, review points, escalation paths and whether the automation still deserves its cost, attention and access.

Shared workflow

See where the work changes hands.

The useful pattern is not one prompt followed by one check. The person and AI shape the work at different moments.

  1. 01 · person

    Define the routine and its owner

    Choose one narrow recurring outcome, name the accountable person, identify approved sources and state the condition that should pause or end the routine.

  2. 02 · ai

    Prepare the repeatable first pass

    Projects, schedules, monitors and connectors

    Use a project, scheduled task or connected workflow to gather approved inputs, identify what changed and prepare the smallest useful draft or alert.

  3. 03 · together

    Tune for exceptions and drift

    Review early runs, add explicit exception rules, remove noisy outputs and update the instructions when definitions, systems or team needs change.

  4. 04 · decision

    Choose what may happen without approval

    The owner decides which steps can run automatically, which external actions require confirmation and when an unusual case must be handed to a person.

  5. 05 · verification

    Audit the routine, not only the output

    Check source freshness, permissions, failed runs, ignored alerts, thread drift and whether people still use the result, then pause or redesign work that no longer earns its place.

The decision stays human

Someone must be able to stand behind the result.

The process owner decides the purpose, permissions, approval gates, exception path and stop condition. AI may run the repeated mechanics, but a named person must remain accountable for the routine and its external effects.

Verification required

  • Confirm each run used current, complete and authorised sources rather than relying on stale retained context.
  • Review connector permissions and external actions using the smallest access required for the task.
  • Test failure, missing data, conflicting instructions and exception cases before relying on an unattended routine.
  • Track whether the output is read, acted on and still useful, then pause routines that create noise or hidden risk.
  • Document an owner, escalation route and stop condition that remain visible to the team.

Failure to watch

The main failure is unattended confidence. A recurring workflow can continue producing plausible output after its source, definition, audience or purpose has changed, while responsibility becomes harder to locate because the process appears automatic.

Skill strengthened

Operational design

Used well, the workflow strengthens process thinking, exception handling, permission discipline and the habit of designing automation around accountable human ownership rather than removing people from the loop.

Worked example

Watch the judgement change the draft.

The first pass is useful. The human correction is what makes it specific, appropriate and worth using.

01 · Input

The honest starting point

Weekly project metrics, a decision log, the open risks and a fixed Friday review format.

02 · AI draft

A useful first pass

This week: two metrics moved, one risk grew and three decisions are still open. Draft review pack prepared with two [CHECK] items.

03 · Human correction

Context changes the answer

The project owner clears a risk that was resolved days ago, explains that one metric's definition changed so the trend is not what it looks like, and cancels a reminder the team had stopped reading.

04 · Final result

The version a person can stand behind

A current weekly review with the changed assumptions called out plainly and nothing fired off externally without a look.

Try it on your work

Start with the real material.

Remove anything you cannot share, fill in the brackets and keep the decision and verification steps visible.

  • Do not automate a process that nobody on the team can explain or check.
  • Do not hoard broad, stale or sensitive context just because it is convenient to leave it lying around.
  • Keep separate pieces of work in separate projects, because a common trap is stored context bleeding from one client or matter into another, sometimes even with memory switched off, and the routine has no way to warn you it happened.
  • Know the limits of a self-running schedule before you rely on it: these routines usually run no more than once an hour, hibernate if you stop opening the app, and cannot always reach your uploaded files or custom setups, and because each run piles onto one long thread the original instructions can get buried over time.

This recurring workflow is for [goal]. Use only [approved sources]. Each cycle: 1) identify what changed, 2) connect it to earlier decisions, 3) list unresolved questions, and 4) prepare the smallest useful next output. Do not assume old information is still true when newer material conflicts. Stop before any external action and ask me to review [checkpoint].

Tools that can support this

Pick the tool after the job is clear.

These are evidence backed starting points, not universal winners. Open a guide or ask Bit to help you weigh the fit.

ChatGPT

Its reviewed guide includes Projects for holding repeated context and Scheduled Tasks for reminders that run on their own.

inferred · moderate confidenceSee why it may fit →Claude

Its reviewed Cowork guidance covers longer-running and scheduled work that still needs an explicit review.

inferred · moderate confidenceSee why it may fit →

Keep going

Build the next useful skill

Reviewed
Review by
Owner
The Human Bit editorial

The Human Bit Weekly

The useful changes, not every launch.

One short issue every Monday. What changed, what it means for your work, and the part that stays yours.

The Human Bit records when and how consent was given. Subscription is confirmed only after the email provider accepts the request.