Week 4
Week 5 of 6
Week 6 Exit Slides

Week 5: AI Advanced

Agents and automation for autonomous operations

The critical distinction

AI assistance requires your active involvement for every task.

AI automation runs independently while you sleep. This week we move from assisted to autonomous.

What AI agents do

Software that performs human-like actions on websites and applications.

Navigate interfaces, make decisions, click buttons, fill forms, complete multi-step processes.

Core agent capabilities

Planning: Given an end goal, agents map out required steps

Actions: Navigate pages, click elements, fill forms, adjust when encountering unexpected situations

Reading and research: Analyze unpredictable web content and make contextual decisions

What this enables

  • Extract structured data from websites
  • Navigate and interact with any web interface
  • Make contextual decisions about next steps
  • Handle variation and edge cases intelligently

The work that used to take hours now runs automatically.

Real use case: Lead generation

  • Research competitor customer lists
  • Gather business information
  • Identify decision makers from LinkedIn
  • Extract contact information
  • Format into structured lead list
  • Automatically add to your CRM

This happens overnight, every night.

Comparing AI agent tools

Claude for Chrome: Research requiring judgment and context

OpenAI Operator: Multi-step processes like bookings

Bardeen: Pre-recorded workflows for LinkedIn and CRM

Browse AI: Scheduled monitoring without supervision

Choosing the right tool

  • Need intelligent judgment? Claude for Chrome
  • Complex multi-step workflows? OpenAI Operator
  • Repetitive template tasks? Bardeen or Browse AI

Match the tool to the task complexity.

Generally acceptable scraping

  • Public business contact information
  • Published pricing and product details
  • Company information from about pages
  • Your own data from platforms you use

If it is published publicly, usually fair game.

Avoid scraping

  • Personal data without consent
  • Content behind paywalls or login
  • Ignoring explicit terms of service
  • Aggressive scraping degrading site performance

When in doubt, check robots.txt and terms of service.

Workflow automation

Middleware platforms let you create automated workflows connecting different apps.

The bridge between systems that do not talk natively.

Simple workflow example

Trigger: New form submission

Automated sequence:

  • Add to Google Sheets
  • AI drafts personalized response
  • Send email via Gmail
  • Create CRM follow-up task

All happens 24/7. Zero manual work.

How workflows work

  • Sequential processing
  • Parallel execution
  • Conditional logic

The power comes from combining these patterns with AI steps.

Comparing workflow platforms

Zapier: Easiest, most expensive

Make: Moderate complexity, better value

n8n: Steeper learning curve, most cost-effective, self-hosting option

Platform selection guidance

Start with Zapier if you value simplicity and speed.

Move to Make when you need sophisticated logic at lower cost.

Consider n8n for maximum control and custom integrations.

Use workflows when

  • Task is repetitive and predictable
  • Needs to run 24/7 without supervision
  • Exact same steps every time
  • Triggering events are clear

Example: Every order triggers identical fulfillment sequence.

Use MCP when

  • Task requires analysis or judgment
  • Asking questions about unstructured information
  • Approach varies based on context
  • You remain involved in the conversation

Example: “Which customers should I prioritize this week and why?”

Use AI agents when

  • Interacting with websites through clicking and typing
  • Extracting structured data from unstructured web pages
  • Navigating complex interfaces without APIs
  • Research requiring decisions based on findings

Example: Build lead lists by researching competitor pages and LinkedIn.

Planning effectively

Give the agent the end goal and specific requirements.

Let it plan the approach. Be specific about what you want, not how to do it.

Action execution

Focus requests on one page or task at a time.

Agents re-plan when exploring unfamiliar territory. Test simple tasks before building complex workflows.

Data handling

Define exact output structure before starting.

Specify spreadsheet columns in advance. Provide examples of expected format. Clarify how to handle missing information.

Starting with workflows

  • Pick one genuinely repetitive weekly process
  • Map out every step in detail
  • Identify what triggers the process
  • Build basic workflow in Zapier or Make
  • Add AI step for personalization
  • Test thoroughly with real data

Starting with agents

  • Identify a research or data gathering task
  • Do it manually once first
  • Use Claude for Chrome or similar to automate
  • Refine prompts based on results
  • Document what works
  • Scale to more complex tasks

Common pitfalls to avoid

  • Over-automating too early
  • Skipping error handling
  • Ignoring monitoring
  • Not documenting

Six months later you will forget how you built this. Document while fresh.

The core takeaway

The progression from assisted to autonomous:

  • Start with workflows for predictable processes
  • Add MCP for intelligent analysis
  • Use agents for web-based research

Each tool solves different problems. Using the wrong tool creates frustration.

Workshop

What you will create:

  • List of 5+ processes you do manually
  • Each classified as workflow, agent, or MCP
  • One fully designed workflow with AI integration
  • Hands-on experience with one agent tool
  • Documentation of lessons learned

Questions?

Next week: AI vs Hiring - When to automate versus when to hire people