Capstone — AI Employee Day

You Made It Here

The image above is a map of everything you have built. Folder architecture. Connectors with permission models. Skills that encode your best workflows. Plugins that bundle skills into deployable units. Scheduled tasks that run while your machine is on. Projects with persistent memory. Content pipelines. Business automation. Dispatch for mobile control. Computer Use for native apps. Sub-agents for parallel workloads. Safety layers. And for some of you, enterprise monitoring.

This module is where you use all of it — deliberately, systematically, on your actual work.

What AI Employee Day Is

The concept is straightforward. Block one full day on your calendar. List every recurring task in your work. Score each one. Build automation for the top three. Repeat quarterly.

The insight Jack Roberts discovered — and that resonates across the community — is that most professionals have between 15 and 30 recurring tasks they do every week. When you actually list them, you find that the majority are rule-based, repeat with little variation, and do not require high-stakes human judgment in their execution. They are just tasks that happen to land on your calendar instead of on an AI's.

"Clear your calendar for one day. List every recurring task. Prioritize by frequency times time. Automate the top three." — Jack Roberts

This is not a one-time event. It is a quarterly practice. Your work evolves, new recurring patterns emerge, old ones disappear. The audit is the discipline that keeps your automation stack aligned with how your work actually looks right now.

Phase 1: The Full Audit

List every task you do that meets three criteria:

• Repeated — you have done it more than three times
• Rule-based — the process does not change significantly each time
• Not requiring high-stakes judgment — the execution is known; decisions stay with you

Do not filter as you list. Write down everything that comes to mind, even if it seems too complex to automate. Prioritize later.

Score each task on three dimensions, each 1–5:

• Frequency — 1 (monthly) to 5 (multiple times daily)
• Time per instance — 1 (under 5 minutes) to 5 (over 1 hour)
• Automation potential — 1 (requires deep judgment throughout) to 5 (entirely rule-based and repeatable)

Multiply all three. Sort descending. Your highest-scoring tasks are your automation targets. The math is not precise — what matters is that it forces you to compare tasks you otherwise would not rank against each other.

Phase 2: The Automation Pipeline Applied

Take your top-scored task through the complete pipeline that has run as a through-line in this course:

1. Do it manually with Co-Work. Not alone. Work through the task together with Co-Work in a project workspace, watching how it handles each step. Note where it needs guidance, where it goes off-track, where it performs better than you expected.
2. Refine until output is consistently good. Run it two or three more times with different inputs. Adjust your instructions based on what you observed. Do not encode until the output is reliably correct — Jack Roberts' rule: "never encode a bad skill."
3. Create the skill. Turn the refined workflow into a skill. Include specific examples of good output if the task has voice or format requirements.
4. Schedule the skill (if it runs on a schedule). Choose the appropriate frequency from the five options: hourly, daily, weekly, on weekdays, or manually triggered.
5. Verify after the first scheduled run. Check the task history. Did it run at the right time? Is the output correct? If not, refine before the next run.

Apply the Three-Test Framework (Jack Roberts) to decide if this task warrants a full plugin:

• Repeat Test: Do you do this more than once a week?
• Chain Test: Does it involve three or more steps across two or more tools?
• Handover Test: Could someone else run this workflow without guidance from you?

If two of three pass, build a plugin. If only one or zero pass, a skill is sufficient.

Phase 3: The Factorio Pattern

The game Factorio starts with manual mining. You automate the mining step, then the processing step, then the assembly step. Each automation enables the next. Eventually the factory runs itself — and you spend your time building extensions to the factory rather than operating it.

The Co-Work equivalent: start with one low-stakes task automated reliably. Then add one connector. Then add one scheduled task. Confirm each new layer works before adding the next. The anti-pattern is trying to automate everything at once before any layer is verified. You get fast, confident, incorrect automation — which is worse than no automation because it takes time to debug and erodes trust in the system.

The progression for this capstone: one solid skill → one scheduled automation → one verified connector integration → one tested plugin. That is a factory. You can build on it next quarter.

Durable Patterns vs. Feature Walkthroughs

Felix Rieseberg (Anthropic) noted that Co-Work ships new features nearly every week. Some of what you learned in specific modules — exact UI paths, feature names, specific configuration steps — will change as the product evolves. The six durable patterns will not.

These are the patterns worth holding onto regardless of what the product looks like in a year:

1. The Automation Pipeline — do it manually → refine → create the skill → schedule it. This is the architecture, not the implementation. It will remain valid as long as Co-Work has skills and scheduled tasks.
2. The delegation mindset — ask "what judgment does this task actually require?" Most tasks require less judgment in their execution than we assume. The question clarifies what can be delegated and what must stay with you.
3. Safety-first design — guardrails are features, not obstacles. The time spent setting up the no-delete guardrail, the check-before-irreversible guardrail, and the anti-injection guardrail is time bought back every time a workflow runs correctly instead of causing a problem.
4. Folder specificity — better context produces better output, regardless of the UI. The principle of selecting the most specific subfolder, writing targeted folder instructions, and giving Co-Work precise scope does not change as the product evolves.
5. Skill quality principle — never encode a bad skill. The cost of running a bad skill on a schedule for a week is higher than the time it takes to refine the skill to quality before encoding. This is true regardless of how skill creation works in future versions.
6. Human-in-the-loop for irreversible actions — this is permanent. Computer Use evolves. Connectors become more powerful. Sub-agents become more autonomous. The principle that irreversible actions require human review becomes more important as capability increases, not less.

"Taste and human judgment remain the key differentiator in the AI-native economy." — Jenny, Anthropic

The capabilities automate execution. Judgment, taste, and strategy remain human. That is not a limitation of the current technology — it is a durable truth about what AI tools are for.

Your 90-Day Optimization Roadmap

The course ends here. Your practice does not.

Month 1: Stabilize. Confirm that the automations you have built are running correctly and producing quality output. Fix anything that is fragile. Document the skills and scheduled tasks you are actively using. Delete or disable anything that is not earning its place.

Month 2: Expand one tier. Choose one capability area from this course that you have not fully implemented and add it. If you are on Tier A/B, add your first plugin. If you are on Tier B, add your first sub-agent pipeline. If you are on Tier C, set up OTel monitoring or organization-wide plugin distribution.

Month 3 onwards: Quarterly AI Employee Days. Block one day per quarter. Run the audit. Score the new recurring tasks that have emerged since the last audit. Run the top two or three through the automation pipeline. Your automation stack should grow by two to three solid skills each quarter.

The compounding effect: each quarter's additions make the next quarter's work faster. The automation stack is a multiplier. It builds on itself if you maintain it with the weekly orchestrator review, the twice-a-week skill check, and the quarterly audit.