How to Automate Almost Anything With Claude (in 20 minutes)

How to Automate Your Work with Claude

Introduction to Automation in Claude

• The video promises to save users at least 30 hours a month by automating tasks with Claude, highlighting the initial time spent managing it manually.

• Emphasizes that there is no one-size-fits-all approach to automation; different levels of automation exist for various tasks.

• The presenter will guide viewers through five levels of automation in Claude, providing real examples for each level.

Level One: Skills

• Skills are defined as pre-canned prompts that can be reused without starting from scratch each time.

• Demonstrates building a skill for company research, which pulls data from LinkedIn and other sources to create a one-pager with talking points.

• Users can create skills using the command slash skill creator, which prompts questions about desired sources and formats.

Building and Managing Skills

• A skill consists of a markdown file containing instructions and connectors; it can also generate scripts if needed.

• Once created, skills are saved in the user's account and can be triggered via slash commands or automatically based on context.

• Skills are ideal for repeatable tasks but require manual activation each time, limiting full automation.

Level Two: Scheduled Automation

• Introduces scheduling as the next level of automation where skills can run automatically at set times without user intervention.

• Scheduling options are available in both Claude co-work and code desktop apps; users choose based on their interface preference.

Example Use Case for Scheduled Tasks

• The presenter plans to automate checking LinkedIn profile views every morning at 7:00 AM, sending connection requests along with personalized messages.

How to Schedule Tasks in Claude Co-Work

Setting Up Scheduled Tasks

• Users can create scheduled tasks by accessing the schedule tab in Claude co-work, where they can select the frequency of the task (hourly, daily, weekdays, or weekly).

• After setting a task to run daily at 9:00 a.m., users must press save to confirm the scheduling. The "run now" button allows for immediate execution and permission acceptance.

Monitoring Task Execution

• Once executed, Claude connects with LinkedIn to check profile views and performs actions like taking screenshots and scrolling through pages.

• Desktop tasks are ideal for personal automation that requires local resources (e.g., logged-in browser sessions), while level two tasks utilize your machine's capabilities without needing API keys.

Transitioning from Desktop to Cloud Automation

• For team-wide automation or when devices may be inactive, level three cloud routines allow tasks to run independently of user machines. This transition involves using connectors and scrapers instead of direct browser access.

Utilizing Connectors and GitHub Repositories

• In cloud routines, skills and context are stored in a GitHub repository which Claude accesses during routine execution. This setup enables efficient management of skills and reference documents.

Creating Cloud Routines

• Users can create new cloud routines via the code tab in the desktop app. A practical example includes scraping blog posts from Anthropic’s website daily and sending updates via Slack.

• The routine named "marketing intelligence" is designed to scrape new blog posts within 24 hours and send notifications to a specified Slack channel.

Configuring Permissions and Environment Variables

• When setting up environments for cloud routines, users should define permissions carefully; full network access may be necessary depending on task requirements.

• It is crucial not to include sensitive information such as secrets or credentials directly into environment variables; instead, use designated connectors for secure handling.

Creating and Managing Cloud Routines

Setting Up the Environment

• The speaker demonstrates how to create an environment for running scripts, such as installing packages like YouTube-DL or yt-dlp.

• A scheduled trigger is set up for 9:00 a.m. daily, with options for full cron customization available.

Selecting Connectors

• The speaker emphasizes best practices by limiting the connectors loaded into the session to only those necessary (Firecrawl and Slack), avoiding unnecessary tools.

Running Cloud Routines

• The routine runs in the cloud independently of the user's laptop, completing tasks even if the device is turned off.

• Two new blog posts from Anthropic are added to the user's second brain, showcasing successful automation.

Advantages of Cloud Routines

• Cloud routines are ideal for tasks that serve multiple users (e.g., team digests or reports), ensuring consistent execution regardless of individual machine status.

• They should be used when tasks require public APIs but not local files or logged-in browsers.

Transitioning to Level Four Automation

• Level four automation allows routines to react instantly rather than on a schedule, enhancing responsiveness in workflows.

• This level requires setting up triggers based on events (like webhooks), making it more dynamic compared to standard scheduled routines.

Configuring API Triggers

• An example routine receives webhooks from Fireflies post-meeting, extracting action items and sending notifications via Slack.

• Specific requirements must be met for webhooks; they need an API token and headers that may not be natively supported by some services like Fireflies.

Using Intermediaries for Webhook Configuration

• To meet webhook requirements, intermediaries like Make may be necessary to format data correctly before triggering Claude routines.

• Once configured properly, these automated processes can seamlessly handle meeting transcriptions and follow-ups through Slack messages.

How to Set Up and Use Claude API Routines

Understanding API Tokens and Endpoints

• The API token is unique and only visible once; it must be copied immediately for future use.

• To obtain the full structure of an API request, access the edit screen and select "call via API" to view example curl commands.

Sending Data to Claude

• The webhook sent to the Claude routine requires a specific format; it cannot process standard JSON but needs data in a designated text field.

• Middleware platforms like make.com are essential for relaying webhooks from systems like Fireflies to Anthropic's URL, ensuring proper headers and credentials are included.

Event-Driven Automation with Claude

• Using an API routine is ideal for event-driven tasks (e.g., call completions or form submissions), as waiting for scheduled runs can delay important follow-ups.

• If a source system cannot send webhooks, consider simpler setups instead of using cloud APIs, which may complicate automation processes.

Levels of Automation with Claude

• Level five automation involves creating applications that utilize Claude for others (teams/customers), differing from levels one through four focused on personal work automation.

• Managed agents provide infrastructure hosted by Anthropic, allowing developers to build apps without managing backend complexities themselves.

Choosing Between Routines and Managed Agents

• For most users automating their own workflows, cloud routines suffice; managed agents cater more towards app development requiring AI integration.

• Managed agents are beneficial in large organizations where tools can be linked to organizational accounts rather than individual ones.

When Not to Use AI in Workflows

• Not all workflows should transition into using Claude; predictable tasks that require consistency (like app syncing or invoice processing) should remain outside AI's scope.

• Automating high-frequency tasks through AI could lead to increased costs and delays compared to traditional methods.

Workflow Optimization with Claude

Understanding Workflow Needs

• A simple rule for workflow management: If your tasks require judgment, writing, or research—especially when dealing with messy data—utilize Claude.

• For straightforward tasks that involve direct data transfer between applications without any cognitive processing, tools like n8n or Make.com are sufficient.

• Most businesses adopt a hybrid approach where complex tasks (like sales prep and follow-ups) are handled in Claude, while simpler app integrations remain in n8n or Make.com.

Addressing Common Misconceptions

• Caution against oversimplification: Claims suggesting to cancel all existing systems overlook the nuances of real-world workflows.

• Introduction of the speaker: Brad specializes in building production systems and offers guidance on implementing these strategies effectively.

• Invitation for personalized assistance: Brad provides AI strategy sessions to help individuals determine which tools best suit their specific workflows, emphasizing tailored solutions.