Make Your Employee Proactive

In the Bronze tier, your employee waited for you to send a message before doing anything. Every real employee you have ever worked with eventually learned to anticipate: checking things before you ask, flagging problems before they become urgent, preparing information you will need before you need it.

This is the difference between an assistant and an employee. An assistant responds. An employee monitors, anticipates, and acts. The shift from reactive to proactive is what makes your AI employee genuinely useful rather than merely convenient.

Your challenge is to identify the single recurring task that would save you the most time if it happened automatically, then make your employee do it on a schedule.

The Challenge

Design and deploy at least one scheduled task that runs on a recurring basis (daily, weekly, or custom interval), produces domain-relevant output, and delivers it through one of your configured channels.

Acceptance Criteria

1. At least one scheduled task is running and fires at the configured interval
2. The task produces output that is genuinely relevant to your professional domain, not placeholder text or generic summaries
3. The output uses real data from your work context (files, sources, or domain knowledge from your skill)
4. You can explain why you chose this specific schedule interval and how it fits your actual work rhythm

Deliverable

Create a scheduler-config.md file in your repo documenting:

• What the task monitors or produces
• The schedule interval and why you chose it
• A sample output from at least one execution
• Your assessment of whether this output would actually save you time

Use Case Gallery

Profession	Scheduled Task	Interval	What It Delivers
Accountant	Scan expense folder for new receipts	Daily, 8 AM	List of new items with amounts, flagging anything over $500 for manual review
Teacher	Compile lesson prep status for the week	Monday, 7 AM	Summary of which lessons are ready, which have gaps, and suggested priorities
Consultant	Summarize client project progress	Friday, 5 PM	Per-client status with hours spent, milestones hit, and overdue items highlighted
Doctor	Check patient follow-up schedule	Daily, 7 AM	List of patients due for follow-up, sorted by urgency, with days since last contact

The best scheduled task is one you currently do manually on a regular basis. If you find yourself checking the same thing every morning or preparing the same report every Friday, that is your candidate.

Cron Is a Polling Loop: The Nervous System Is What Comes Next

What you are building here is a polling loop: NanoClaw checks a schedule and fires a task at the configured interval. This is a valid and useful starting point.

The production version is different. The thesis calls it Invariant 7: the nervous system. In a full Agent Factory, work arrives via external triggers (webhooks, customer events, API calls), Workers hand off to other Workers via internal events, and a durable execution layer (Inngest) survives crashes mid-flow and shapes traffic under load. A cron job cannot do any of that.

You will build the full nervous system in a later part. For now, understand that the scheduler you configure here is the proto version: it answers the question "can this employee act without being asked?" The answer is yes. The production architecture that makes that reliable at scale comes later.

Hints

PRIMM-AI+ Practice

This challenge follows the PRIMM-AI+ cycle. Before you build, predict. After you build, investigate.

Predict [AI-FREE]

Before you configure any scheduled task, write down:

• What data source your task will monitor, what interval you will choose, and what the first delivered output will say (write a rough draft of the first 3 bullet points).
• Whether the output will feel genuinely useful or like a placeholder summary when you read it cold on a Monday morning.
• What will go wrong on the first test run (wrong format, missing data, wrong channel, fires at unexpected time).
• Your confidence score from 1 to 5.

Do not ask the agent until those notes are written.

Run

Use Claude Code ($ claude) to design the scheduled task logic and draft the NanoClaw scheduler configuration. Set the interval to a short value (every few minutes) for the first test run, verify the output, then switch to your production schedule.

Complete the challenge as described above: deploy the scheduled task, verify it fires at the configured interval, confirm the output uses real domain data, and document everything in scheduler-config.md.

Investigate

First, write your own explanation of whether the delivered output matched the rough draft you wrote in Predict (both in content and usefulness). Then ask the agent: "What is missing from this scheduled output that a human professional would have included in the same report?"

Modify

Change one requirement: adjust the schedule interval to one step shorter or longer (for example, from daily to every 12 hours, or from weekly to daily) and predict how the volume and specificity of the output will change. Update the configuration, let at least two runs fire, compare the outputs, and assess whether the new cadence is better or worse for your work rhythm.

Make [Mastery Gate]

Verify against the acceptance criteria above. Passing means: at least one scheduled task fires at the configured interval, the output is domain-relevant (not placeholder text), the output draws on real work-context data, and scheduler-config.md documents the task design with a sample output and your assessment of its value.

Level 1: Where to Look

NanoClaw supports scheduled tasks through its task management system. Check the /setup documentation and the NanoClaw repository for how to configure recurring jobs. Look at how existing scheduled tasks are defined: the pattern is: specify WHAT to do, WHEN to do it, and WHERE to send the output.

Level 2: Ask Your AI

Send this to Claude or your AI employee:

"What recurring task would save me the most time if automated? I'm a [your profession]. Consider my work rhythm: when during the day or week do I most need information delivered proactively? What data source would this task need to monitor?"

Use the answer to narrow your choice before building.

Level 3: Design Framework

Break your scheduled task into four decisions:

1. WHAT to monitor: The data source or condition your employee checks (a folder, a calendar, a list, a status)
2. WHEN to check: The interval that matches your work rhythm (daily at 8 AM? Monday mornings? Every Friday evening?)
3. WHAT to report: The output format: what information and how it is structured (bullet list? table? prioritized flags?)
4. WHERE to deliver: Which channel receives the output (WhatsApp? Slack? Gmail? The channel from Connect Your Employee to the World?)

Configure the task in NanoClaw using its scheduling capabilities. Start with a short interval for testing (every few minutes), verify the output looks right, then switch to the production schedule.