Recurring Agent Worker Conversion
Turn a validated workflow into an always-on skill with exception-based check-ins
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 6
- Confidence
- 98%
Recurring Agent Worker Conversion transforms a successful interactive workflow into a proactive operating unit. After the task has been validated, its instructions, context, connector use, output format, and review gates are saved as a skill. A scheduler launches that skill at the required cadence and can incorporate results from prior runs, such as event performance or attribution data. Instead of rebuilding the project, the human receives a prepared result, approves it, and handles exceptions. Notifications can arrive through the team's normal environment, including Slack or email. The mechanism shifts work from reactive prompting to persistent execution while preserving change controls. Over time, better models, broader connectors, and added organizational context can improve the worker, but its assumptions and outputs still require periodic review.
Origin
Nate Folin described converting the executive-dinner workflow into a weekly Perplexity Computer skill and extending the pattern across recurring RevOps work on Marketing Against The Grain.
Core principles
- 01Convert repeated work into a reusable operating asset
- 02Schedule stable workflows instead of restarting them manually
- 03Reduce human work to approvals and exceptions
- 04Feed performance from prior runs into the next run
- 05Let recurring agents report outcomes through existing channels
How to run it
- 1
Select a repeated flow
Choose a workflow that recurs and has already passed stage-by-stage validation.
Pro tip Start with work that consumes substantial setup time but follows stable logic.
Watch out Do not schedule a workflow that still requires frequent improvisation.
- 2
Encode the skill
Save the procedure, required context, tools, decision rules, and expected outputs as a reusable skill.
Pro tip Include examples of acceptable and unacceptable results.
Watch out Missing context will force the agent to reconstruct assumptions on every run.
- 3
Set cadence and inputs
Schedule the skill and define which fresh data and prior-run results it should inspect.
Pro tip Align the schedule with the availability of complete source data.
Watch out Running before source systems update can produce misleading conclusions.
- 4
Define intervention points
Specify which results can proceed automatically and which require approval or review.
Pro tip Route low-confidence cases into a manual-review queue.
Watch out Silently forcing ambiguous cases through the happy path can corrupt downstream systems.
- 5
Report completion
Send a compact status message containing the prepared output, performance indicators, and exceptions.
Pro tip Link directly to the underlying artifact or queued action.
Watch out A vague success notification hides partial failures.
- 6
Maintain the worker
Periodically revisit the skill as teams, systems, and business priorities change.
Pro tip Use observed exceptions to improve instructions and validation rules.
Watch out An unattended skill can preserve obsolete assumptions indefinitely.
In the wild
Every Monday, an agent reviews prior event results and current opportunities, chooses a prospective city, prepares the audience and campaign, and alerts the events team that everything is queued. The team spends its time checking and approving the proposal instead of reconstructing the project.
→ The recurring project falls to a roughly 30-minute weekly check-in.
Separate workers inspect attribution, schedule proactive meetings, or alert a finance lead when a sufficiently large finance prospect books a sales call. Each worker monitors a narrow condition and reports the relevant event.
→ Small operational requests no longer require intake meetings and manually implemented reports.
Common mistakes
Scheduling before validation
A recurring schedule multiplies defects in an unproven workflow.
No exception path
The worker needs a clear way to surface ambiguity instead of guessing or stopping silently.
Never revisiting assumptions
Organizational, model, connector, and data changes can make a once-reliable skill obsolete.
Is it for you?
Best for
Stable weekly or monthly workflows with consistent inputs, outputs, and review boundaries.
Not ideal for
It is unsuitable for novel projects whose objectives or procedures change substantially on every run.
From the transcript
“we we translate this over to a skill and we can run a weekly job or on this skill”
“But once this is always on, it turns into a 30-minute check-in of like, oh, let's just approve this each each week versus let's do…”
“you've gone from like reactive agent who's doing things as you're talking to it to a proactive agent that's just doing work for you”
From the episode
How 1 Human + AI Replaced a 15-Person RevOps Team