Automating Work
Automating Work
Section titled “Automating Work”Interactive Claude is already powerful — you ask, it does. But there’s a next level: Claude that does things without you asking.
That’s what this section is about. Running Claude on a schedule. Running it in the background while you do other things. Running multiple Claude instances in parallel, each handling a different piece of a bigger problem.
The progression
Section titled “The progression”Most people move through these stages:
Stage 1: Interactive — you’re in the conversation, asking questions and reviewing answers. This is where everyone starts, and honestly where most work happens.
Stage 2: Headless — Claude runs from a script or scheduled job without a conversation interface. You give it a task and walk away. Results go to a file, an email, a Slack message, wherever you want them.
Stage 3: Multi-agent — a coordinating Claude instance (the orchestrator) breaks a big job into pieces and delegates each piece to specialized sub-agents running in parallel. Used for large, complex tasks.
You don’t need Stage 3 to get real value from automation. Stage 2 alone — a Claude job that runs every morning and gives you a briefing — is genuinely useful and not that hard to set up.
When automation makes sense
Section titled “When automation makes sense”Automation is worth setting up when a task is:
- Recurring — you do it daily, weekly, or on some regular schedule
- Predictable — the inputs and expected outputs are consistent
- Low-stakes if wrong — or has a review step before anything permanent happens
- Time-consuming — enough that doing it manually is a real cost
Interactive is better when a task is:
- One-off — you’ll never do this exact thing again
- Exploratory — you’re figuring out what you even want as you go
- High-judgment — the right answer depends on nuance that’s hard to encode upfront
!!! tip “Don’t automate before you’ve done it manually a few times” The best automations come from tasks you’ve already done interactively and know well. If you don’t understand the task yourself, you can’t write a prompt that handles it reliably.
What’s in this section
Section titled “What’s in this section”| Page | What you’ll learn |
|---|---|
| AI That Works While You Sleep | Headless mode, cron scheduling, practical automation examples |
| AI Teams: Multiple Agents | Sub-agents, parallel work, the orchestrator pattern |
Start with headless and cron — it’s where the practical wins are. Multi-agent is powerful but more complex; read it when you’re comfortable with the basics.