⁠6 EASY Steps to Automate 99% of Your Work With Claude

Summary

Austin Marchese presents a six-step framework for building practical, maintainable automations with Claude. His central point is that an automation should be easy to test and maintain—not merely powerful.

Minimum viable setup

An automation has three parts:

• Trigger: Starts the workflow.
• Logic: Determines what it does.
• Monitoring: Shows what happened and whether anything needs fixing.

Marchese’s example setup uses Claude Routines as triggers, Claude Skills for reusable logic, and the Claude desktop app to monitor and test results. He argues that this simple structure is preferable to complex setups for most users.

The six steps

1. Build a quick prototype (“automation spike”). Start with a rough end-to-end test to confirm the idea is technically possible before polishing it. One example is a daily briefing that gathers email and Slack activity, summarizes what was missed, adds an inspirational quote, and emails the result. Claude’s assurances are not enough: connect the services and run the workflow.

2. Make the automation transferable with a Skill. A Routine can run on a schedule, but repeatedly pasting a long prompt into it is harder to maintain and may be less consistent. Instead, put the reusable instructions in a Claude Skill and have the Routine reference it. Updating and testing the Skill lets the Routine use revised logic without being rewritten. Marchese calls this skills-based automation.

3. Choose when, how often, and where it runs. Schedule an automation for when it will have the greatest practical impact, rather than simply choosing the most obvious time. Run it only as often as needed to reduce noise and usage. Prefer local execution when possible; use the cloud when the machine must reliably run at a particular time or missing a run would be costly. Local runs depend on the computer being active, while cloud execution adds environmental complexity.

4. Improve the automation in four ways.

   • Calibrate: Test the Skill interactively, refine its output, and document pitfalls.
   • Add features: Extend the workflow—for example, by checking a calendar. Plugins or MCP connections can provide integrations and capabilities without requiring everything to be built from scratch.
   • Adjust the output: Choose a format such as plain text, HTML, or a shareable Claude artifact.
   • Improve system design: Create self-correcting workflows where appropriate, and self-interrupting ones that stop on errors or after a defined number of cycles. Clear stop conditions help prevent runaway loops and wasted tokens. In Marchese’s example, the automation checks that required connections work before proceeding; if one is missing, it reports the problem and stops.

5. Design for scale and lower maintenance.

   • Avoid fixing the same issue separately in multiple automations. Create reusable supporting Skills—for example, a shared Skill for sending Slack messages.
   • Send notifications to one place that you regularly check.
   • Make completion reports specific. Instead of saying a task succeeded, report what was actually processed, such as the number of emails, messages, or calendar events reviewed. This helps catch what Marchese calls “green light drift”: an AI claiming success despite incomplete work.

6. Automate only what is valuable. Prioritize workflows that remove a real bottleneck or improve your ability to create value. Reassess and retire automations that no longer help, keeping useful information high and noise low.

Examples and product discussion

Marchese describes two automations he uses:

• A Friday HTML digest of important tweets and related stories, which helps him avoid spending time on Twitter.
• A weekly process that gathers information from YouTube, email, Granola, and Slack, then adds relevant updates to a cloud project.

The video also includes a CodeRabbit sponsor segment. Marchese describes it as a code-review tool that reviews pull requests, identifies issues, suggests and can apply fixes, summarizes changes, and supports questions through a chat agent.

Main speakers and sources

• Austin Marchese: Presenter and source of the automation framework and examples.
• CodeRabbit: Sponsor discussed in a product demonstration.