⚡ Workflows & Automation

Workflows & Recurring Tasks

This is where Context OS and Data OS start paying off. You've built the foundation — now you systematically audit what you're doing manually, identify what can be automated, and start freeing up bandwidth to work on the business instead of in it.

📽 Workshop recording
~30 min read
Builds on Setup, Context OS & Data OS
Real examples from a live system
Section 01

The Bandwidth Problem — Working IN vs. ON the Business

Most founders are running at 10–20% available bandwidth for strategic work. The rest is consumed by operational tasks — filming, scripting, handling support, writing proposals, taking sales calls, managing deliveries. You're working in the business constantly, and you never have enough time to work on it.

The cliché is true: spending all your time keeping the flow going means you're not doing anything to actually move forward. And then you wonder why things aren't growing. The goal of everything we're building is to shrink that in-business load down as far as possible.

Before AIOS
80–90% — Working IN the business
10–20% — Working ON the business
Content, support, proposals, delivery, admin. Day after day. No room to build.
After AIOS
~20% — Minimum unavoidable in-business
~80% — Working ON the business
New channels. New offers. New systems. New hires. The work that actually grows things.
💡
The real payoff: When you finally free up this bandwidth, you don't just get time back — you have the most powerful AI business assistant you've ever had, with full context of everything, ready to apply full force to whatever you put in front of it. All that freed time, plus an AI co-CEO that knows your business inside out. That's when the compounding starts.
Section 02

The Task Audit — Map What You Currently Do

Before you build a single automation, get everything you currently do out onto paper. Not what you could do, not what you want to do — what you are actually doing right now, day to day, that takes up your time.

⚠️
Scope discipline is critical. When you do this exercise, you'll get excited and start adding things that aren't on your current plate — "I should build a tool for my sales reps," "I should set up a new content channel." Stop. Those go on a separate initiatives list for when you have freed bandwidth. The audit is only for tasks you are currently doing.

Here's an example of what a real task audit looks like across a business with multiple functions. Map your own version before you decide what to automate:

Content
  • Content ideation and planning
  • Scripting and filming scheduling
  • Repurposing across platforms
  • Publishing and distribution
  • Performance monitoring
Agency / Client Work
  • Discovery call prep and follow-up
  • Proposal writing and scoping
  • Project management and status
  • Technical delivery review
  • Funnel analytics and reporting
Education / Community
  • Workshop planning and prep
  • Module writing and production
  • Community monitoring and response
  • Student progress analysis
Operations
  • Data review and business reporting
  • Feature prioritization (SaaS)
  • Team communication and strategy
  • Sales pipeline management

Once you have your list, sort each task into: augment (AI helps me do it faster) or automate (AI does it without me). The distinction determines which type of workflow to build.

Section 03

Two Types of Workflows

Every task on your list falls into one of two buckets. Most things start as augmented and graduate to automated once you've refined them enough to trust the output without being in the loop.

🤝
Type 1
Augmented Workflows
You and AI work together — faster and better than you alone. You invoke a slash command, the workflow guides you through a structured process, AI does the heavy lifting, you provide direction and final judgment.
→ Built with slash commands
⚙️
Type 2
Automated Workflows
AI runs the task without you — on a schedule or triggered by an event. You review outputs rather than run the process. Works for tasks that are well-defined, low-risk, and where you've already dialed in the quality through augmented use.
→ Built with cron jobs or triggers
🔄
The natural upgrade path: Most tasks start as augmented. You run the workflow yourself, refine it, dial in the quality. Once you trust it — once you stop correcting the same things each time — you graduate it to automated. Never try to automate something you haven't first run in augmented mode many times.
Section 04

Augmented Workflows — Slash Commands

Augmented workflows live in your .claude/commands/ folder as markdown files. You invoke them with a slash prefix. They walk you through a structured process — asking questions, pulling context, doing research, producing outputs — while keeping you in the loop for judgment calls.

🏭
Real example — /course-module (built in a single morning): A workflow for creating new course content. It reads all the accelerator context, pulls from the database, asks whether this should be a mini / standard / full module, analyzes all dumped materials, and walks you through planning the content — turning a month-long manual process into a day.
How a slash command workflow is structured
1
Read context
Load the relevant context files — business context, course content, agent status, any databases relevant to this workflow.
2
Pull and analyze materials
Gather all inputs — workshop transcripts, community questions, YouTube comments, competitor content — whatever feeds this workflow.
3
Shape and scope — chat back and forth
Ask clarifying questions. Is this mini / standard / full? What angle? What audience? Get clear before doing any heavy lifting.
4
Execute with your input
Build the output — draft, plan, analysis, content — with you reviewing and directing at key decision points.
5
Output to files
Save the result to the appropriate folder. Plans to plans/, deliverables to outputs/. Always structured so you can find and reference it later.

Any task on your audit list that requires your creative input or judgment — content ideation, proposal writing, strategic planning — is a candidate for an augmented slash command. Build the methodology once, run it many times.

Section 05

Skills vs. Commands — Which to Use

Both are invocable with a slash prefix. The difference is weight and structure. For most workflows, a slash command is the right choice. Skills are for when you need to bundle reference materials, examples, or scripts alongside the workflow itself.

When to use each
Slash Commands Lightweight
  • Single markdown file in .claude/commands/
  • Best for most workflows — structured process, step-by-step instructions
  • Can reference scripts and context files already in the workspace
  • Quick to build, easy to iterate on
  • The default choice — prefer this unless you need more
Skills Heavier
  • Full folder with SKILL.md + references, examples, scripts subdirectories
  • Better when the workflow needs bundled reference materials alongside it
  • Examples: AIOS visual skill (includes design reference library), funnel generation skill (includes templates)
  • Worth the overhead only when you have associated assets to ship with the workflow
Section 06

Automated Workflows — Cron Jobs & Triggers

Once a workflow is refined to the point where you trust the output without being in the loop, you disconnect yourself from it. It runs on a schedule or fires on an event. You review results. Two mechanisms to know:

Type A
Cron Jobs — Time-based
  • Run at a fixed time — daily, weekly, hourly
  • macOS: use Launch daemon (launchd) for reliable scheduling even after restarts
  • Keep laptop plugged in and open overnight so jobs run as scheduled
  • More deterministic and controlled than OpenClaude-style setups
  • Best for: data collection, daily brief, regular audits, weekly planning
Type B
Trigger-based — Event-driven
  • Instant trigger: Something happens → do X immediately (e.g., Telegram message received → process it now)
  • Polling trigger: Check every 15 minutes if something changed — good for non-urgent events like voice memo processing
  • Choose instant when timing matters; polling when eventual processing is fine
  • Best for: file watchers, inbound messages, status changes in external systems
🔧
How to set up a cron job: Just ask Claude Code — "I want to set up a cron job at 6 AM that runs my data collection script." It will configure macOS launchd, write the necessary plist file, and walk you through testing it. You don't need to know launchd syntax. The workflow is: build the slash command first, refine it through augmented use, then wire it to a cron when you're ready to step back.
Section 07

A Real Cron Schedule — What's Actually Running

Here's the actual automated schedule running inside a live business. This is what "work happens while you sleep" looks like in practice. Each job was built as an augmented workflow first, refined over many runs, then wired to a cron.

Time
Job
05:30
Data Collection
Scripts pull from all APIs — YouTube, Google Analytics, Bitly, Stripe, Calendly, Google Sheets. Takes daily snapshots, writes to local database. The Data OS layer running automatically.
06:00
AI Model Tracker (AI Doc Scan)
Pings LM Arena API, compares top models against current workspace docs. If something changed: researches the new model (API docs, use cases, integration snippets), updates the AI reference docs so Claude Code always knows the best model for every task.
06:30
Data Audit
AI agent reviews everything collected. Checks for gaps, errors, anomalies. Fixes issues on the spot. Generates a report of what ran, what broke, and suggestions for system improvement. Data quality gatekeeper.
07:00
Daily Digest
Pulls all collected data + Fireflies meeting intelligence from the past day. Assembles a full business brief: what happened, key metrics, anything unusual. Sends to Telegram + saves PDF to outputs/ for reference.
Sun 08:00
Memory Prune
Claude Code accumulates memory during chats that fills token space. Weekly: exports all memory, compares against existing context docs, extracts anything valuable and updates context files, then wipes the memory clean. Keeps context lean for maximum intelligence per session.
Weekly
Workspace Audit (planned)
Check for trash files, archive old plans and outputs, ensure folder structure is clean and findable. Keeps the workspace organized as it grows. The workspace itself needs hygiene maintenance — build this early before it becomes unwieldy.
📐
The sequencing matters: Data collection runs first (5:30 AM), then the digest (7:00 AM) can pull from fresh data. The audit (6:30 AM) runs between them — catching data errors before they get summarized into the brief. Order your cron jobs to respect dependencies.
Section 08

/explore — From Idea to Built Automation

When you have a new automation idea but aren't sure how to build it or what's technically feasible, /explore is your starting point. It takes you from a rough description through a structured discovery process — producing a full exploration document you can then hand to a fresh implementation agent.

1
Discovery — get clear on what you want
Describe the idea. The command asks clarifying questions: what's the trigger, what's the output, who reviews it, what data does it need? Sharpens the vague idea into a concrete scope.
2
Research — find the best technical approach
Searches the web and your existing workspace. Finds APIs, libraries, patterns that can be used. Checks what you've already built that could be reused. Surfaces options you wouldn't have known to look for.
3
Shaping — define and confirm the feature
Chat back and forth until it's super clear. Confirms the scope, the data sources, the output format, the delivery mechanism (Telegram, file, dashboard, etc.). No surprises in implementation.
4
Full exploration document generated
Produces a complete spec doc — architecture, data flow, technical approach, edge cases, implementation steps. Hand this directly to a fresh primed agent with /implement.
🎙
Voice + /explore = maximum speed. Turn on voice input (Whisper Flow or any voice-to-text), run /explore, and talk through the idea. The discovery phase handles the back-and-forth. You don't need to type out a specification — just describe what you want to build in plain English and let the command structure it into something buildable.
Section 09

Memory & Workspace Health

Two things that become important as your workspace grows: Claude's memory filling up (costing you token space and intelligence per session), and your workspace structure drifting into disorganized territory (making it hard to find things). Both need ongoing maintenance.

  • Claude's memory fills token space. When you run /context, you'll see how many tokens are used by memory vs. tools vs. system. If memory files are bloated, you're giving up intelligence capacity every session. The Sunday memory prune automates the cleanup.
  • Memory prune workflow: Export all Claude memory → compare against existing context docs → extract anything valuable and update the context files → wipe memory clean. Result: fresh memory space, no lost learnings, context docs get richer over time.
  • Workspace audit: As you ship new commands, plans, and outputs, the workspace accumulates clutter — old plans that were superseded, outputs that should be archived, scripts that are no longer used. A weekly workspace audit cron checks for trash, archives stale files, and confirms the structure is still navigable.
  • Keep CLAUDE.md in sync. Every time you add a new command, script, or structural change, the workspace can drift from what CLAUDE.md documents. The implement command should include a documentation step — but spot-check CLAUDE.md regularly to ensure it still reflects reality.
🧹
A workspace that runs your business needs to be maintainable. The more you add, the more important hygiene becomes. Build the workspace audit and memory prune early — before the workspace grows to the point where finding anything requires a search. These are small automations with large quality-of-life payoffs.
Key Takeaways

What to Remember

Takeaway 01
Scope the audit to what you currently do
Write down existing tasks only. Not what you wish you did or could do — what you actually do every day. Everything else is an initiative for later.
Takeaway 02
Augmented first, automated later
Every automation starts as an augmented slash command. Run it yourself. Refine it. Only graduate to automated when you trust the output without being in the loop.
Takeaway 03
Cron and triggers serve different needs
Time-based (cron): data collection, daily brief, weekly audits. Event-based (trigger): inbound messages, file changes, status updates. Choose based on whether timing matters or eventual processing is fine.
Takeaway 04
Keep the machine running overnight
Your cron jobs run while you sleep — but only if the machine stays on. Laptop plugged in, lid open, sleep disabled. Eventually move to a Mac Mini or VPS for always-on reliability.
Takeaway 05
/explore turns ideas into specs
Don't try to spec new automations in your head. Use /explore with voice input to go from rough idea → discovery → research → shaped spec → implementation-ready document. Pair with a fresh /implement agent.
Takeaway 06
Memory prune and workspace audit are non-optional maintenance
Claude's memory fills up and costs token space. Your workspace accumulates clutter. Both degrade performance over time. Automate the cleanup from day one — don't wait until things feel broken.
Homework

Do This Before the Next Session

  • Write your task audit. Get everything you currently do on paper. Every business function, every recurring task, every daily thing that takes time. Don't filter — just list. Aim for 20–40 items. This is your automation backlog.
  • Separate tasks from initiatives. Go through your list and mark each as: "I currently do this" (task → automate) or "I wish I could do this" (initiative → defer). Move all initiatives to a separate list to tackle when you have freed bandwidth.
  • Pick your first augmented slash command. Find the most repetitive task on your list that involves creative input. Use /explore (or /create-plan) to map out what the workflow would look like. Build it. Run it 5 times this week.
  • Wire your data collection to a cron (if you haven't). If you set up data OS connectors from the previous session, get them on a 5:30 AM schedule. Start the snapshot clock now. Every day without it is data you'll never have.
  • Set up the memory prune cron. Ask Claude Code to set up a weekly Sunday memory prune. It will write the workflow and configure the cron. This is a small thing with compounding returns — a clean, lean context every week.
Continue Learning

Related Modules