§ 01
How To Say It
Facilitator Talking Points
Use the transcripts' simple, direct language. The strongest move is to keep returning to the same question: what should happen automatically after the trigger?
Opening · What Is an Automation?
An automation is simple: when something happens, a set of steps runs automatically. It is not magic, and it does not need to be complex. If someone fills out a form, they get a confirmation email, their info goes to a spreadsheet, and you get notified.
01 · Repetition
Do you do it again and again?
Repeated emails, copied data, recurring follow-ups, and routine organizing are strong candidates.
02 · Low Judgment
Does it require little decision-making?
Standard responses, moving data, tagging records, and notifications usually belong in automation.
03 · Friction
Does it feel annoying, slow, or wasteful?
Friction is useful signal. If people avoid the work or forget it, the system should help.
04 · Formula
Frequency x Time x Annoyance
If it happens often, takes time, and annoys you, it is worth evaluating for automation.
Workflow First, Tools Second
The key line for the mapping segment is: if you cannot explain your workflow clearly in plain English, you will not be able to build it. Make the invisible process visible before anyone opens Relay.
01
Trigger
What starts the workflow?
02
Inputs
What data comes in?
03
Steps
What happens in order?
04
Output
What do you get at the end?
Automation vs. AI Workflow vs. Agent
| System Type | Best For | Workshop Example |
|---|---|---|
| Automation | Stable, predictable steps | Form submitted → add row to Google Sheet → send Slack notification. |
| AI Workflow | Understanding or transforming information | Form submitted → AI summarizes the message → AI classifies lead intent → notify with summary. |
| AI Agent | Narrow goals where the path can vary | Research this company, decide whether the lead is a fit, draft a personalized reply, and send the draft for approval. |
Automation moves data. AI interprets data. Agents reason across steps toward a goal.
Say this during the AI and agent segments
Common Misconceptions
"I should automate everything."
No. Automate repetitive, low-judgment, high-friction work. Keep humans involved where judgment, relationship, risk, or taste matters.
"I need to pick the perfect tool before mapping."
No. Start with Trigger, Inputs, Steps, Output. Tools are easier to choose once the workflow is clear.
"Adding AI means the system should make every decision."
No. Start with small AI tasks: summarize, classify, extract, draft. For important work, review the AI output before anything goes out.
"An AI agent is a fully autonomous employee."
No. A practical agent is a flexible workflow with a narrow goal, tool access, instructions, context, and boundaries.
§ 02Demo → Practice
Build Flow
The same lead-capture build, evolving from automation to AI workflow to agent — demo then practice, in the order you'll run them.
Demo 01 · Automation
Open Relay template: lead-capture automationGoal
Show a simple fixed-path automation: form submitted, lead saved, notification sent.
Step-by-Step
- Open the pre-built Relay workflow.
- Point out the form submission trigger.
- Show the Google Sheets action and mapped fields.
- Show the Slack or email notification action.
- Submit one test form and show the Sheet row plus notification.
Facilitator Line
"This is automation: a trigger starts a fixed sequence of steps."
0:43-1:05 · Practice 1
Build Your First Relay Automation
Objective: Create one working automation with one trigger and at least two actions.
- Set the trigger.
- Add a storage action, such as Google Sheets.
- Add a notification action, such as Slack or email.
- Run one successful test.
- Submit what it does plus a screenshot or short description of it working.
Demo 02 · AI Workflow
Open Relay template: lead-capture AI workflowGoal
Show how the fixed automation becomes smarter when AI summarizes and classifies the lead inquiry.
Prompt to Use
TASK: Summarize this lead inquiry in one sentence. Then classify the lead as high intent, medium intent, or low intent.
CRITERIA:
High intent: specific business problem, clear interest, asks for demo or next step.
Medium intent: interested but vague or exploratory.
Low intent: unrelated, unclear, or no business need.
OUTPUT FORMAT:
Summary:
Intent:
Reason:
Step-by-Step
- Open the workflow with the AI step added after the trigger.
- Pass the form message into the prompt.
- Update the Slack/email notification to include summary, intent, and reason.
- Submit a lead with a long message and show the summarized notification.
Facilitator Line
"The automation still handles the structure. The AI handles the flexible thinking inside the structure."
1:10-1:35 · Practice 2
Add One AI Step
Objective: Upgrade the automation by adding one simple AI task.
- Choose summarize, classify, draft, or extract.
- Write a specific prompt for the AI step.
- Use the AI result later in the workflow.
- Test and compare the output before and after the AI step.
Demo 03 · AI Agent
Open Relay template: lead-capture AI agentGoal
Turn the workflow into a narrow agent without making it dangerously broad.
Agent Brief
GOAL: Evaluate inbound leads and draft personalized replies.
TOOLS: Lead form data, Google Sheets, web research, email draft.
INSTRUCTIONS: Research the company, identify the primary business problem, decide whether the lead is high quality, and draft a concise response.
BOUNDARIES: Draft only. Do not send automatically. If information is missing, ask a follow-up question.
OUTPUT: Fit assessment, reason, draft reply, recommended next step.
Step-by-Step
- Start from the AI-enhanced workflow.
- Replace the single classification task with the broader lead-evaluation goal.
- Add only the tools the agent needs.
- Write boundaries before testing.
- Run one test lead and inspect the draft before approval.
1:50-2:00 · Practice 3
Build a Narrow First Agent
Objective: Ship the first version of an agent that does one useful thing well.
- Define the agent goal.
- Choose the tools it can use.
- Write instructions and boundaries.
- Define the output.
- Run one test case and submit how it works, with screenshots or a link if possible.
§ 03
Temperature Reads
Check for Understanding
Use these checks while learners build. The fastest way to fix confusion is before it becomes a broken workflow.
Quick Checks
- "What is the trigger in your workflow?"
- "What data comes in at the start?"
- "Which step is storage, and which step is notification?"
- "What is your AI step doing: summarize, classify, draft, or extract?"
- "Is this still an automation, or does it have a goal and flexible steps?"
- "Where does the human review happen?"
When Learners Are Confused
If they cannot name a trigger
Ask: "What happens first in the real world?" Form submitted, email received, meeting ended, file uploaded, row added. That is usually the trigger.
If they are overbuilding
Reduce the workflow to one path. One trigger, two actions, one output. Complexity comes later.
If their AI prompt is too vague
Ask what output they need and how it should be formatted. Then add criteria. "Analyze this" becomes "summarize in one sentence and classify intent as high, medium, or low using these criteria."
If they want the agent to do too much
Narrow the goal. "Help with sales" becomes "Evaluate inbound leads and draft personalized replies."
§ 04
Keep the Energy
Engagement Strategies
This workshop works best when learners keep applying the same concept to their own work, not only the lead-capture demo.
Start with weekly annoyances
Ask learners to name repetitive work from the last seven days. Recent pain creates better automation ideas than abstract brainstorming.
Use pair critique before build time
Have learners explain their trigger, steps, and output to a neighbor. If the neighbor cannot understand it, the workflow is not ready for Relay.
Make testing public
Celebrate test runs, including broken ones. A failed test is useful if learners can name which field, step, or prompt caused the issue.
Keep returning to the same diagram
Update the lead workflow visually at each major segment. Learners should see exactly what changed when AI and agent behavior were added.
Use "draft, don't send" as the safety mantra
When learners get excited about agents, make the first version produce drafts and recommendations, not irreversible actions.
§ 05
Contingency Plans
What Could Go Wrong
Most problems are caused by unclear workflow thinking, broken field mapping, or oversized agent scope.
Learners start in tools too early
Pause tool use and return to Trigger, Inputs, Steps, Output. Require a plain-English workflow before building.
Relay trigger does not fire
Check that the form was submitted after the automation was enabled. Run a fresh test and use the simplest possible form fields.
Fields map incorrectly
Reduce to name, email, and message. Verify each field one at a time before adding optional fields.
AI output is vague
Add criteria and output format. Do not ask the AI to "analyze"; tell it exactly what categories, fields, or draft structure to produce.
Agent scope gets too broad
Narrow the goal to one job. Good: "Evaluate inbound leads and draft replies." Bad: "Manage my sales pipeline."
Learner wants full autonomy
Move the first version to recommendation or draft mode. Human approval comes before sending, scheduling, deleting, purchasing, or changing records.
Tool access or login fails
Switch to the worksheet version. Learners can still map the workflow, write AI prompts, and scope the agent without live tool access.
§ 06
Anticipated Questions
FAQ & Q&A
Short answers facilitators can use without turning the workshop into a technical lecture.
Automation Questions
"How do I know if something should be automated?"
Look for repetition, low judgment, and friction. If it happens often, takes time, and annoys you, it is a candidate.
"What should not be automated?"
Anything that requires nuanced judgment, trust, personal relationship, high-stakes decision-making, or sensitive data should start with human review.
"What if my workflow has lots of branches?"
Start with one path. Build the most common path first. Add branches only after the basic version works.
"Do I have to use Relay?"
No. Relay is the workshop example. The principles transfer to Zapier, Make, n8n, Airtable automations, or other workflow tools.
AI Workflow Questions
"Where does AI fit best inside an automation?"
Use AI where information needs to be summarized, classified, extracted, drafted, compared, or transformed.
"Can I trust the AI classification?"
Treat it as a useful first pass. For important outcomes, review it or route uncertain cases to a human.
"Why not make the AI do everything?"
Broad AI tasks are less reliable. Simple, narrow AI steps are easier to test, debug, and trust.
Agent Questions
"What makes an agent different from an automation?"
An automation follows predefined steps. An agent has a goal and can decide which steps to take within boundaries.
"Do agents need tools?"
Useful agents usually do. The AI model reasons; tools let it search, read, write, update, draft, or retrieve information.
"What is a good first agent?"
A narrow one: lead qualification, meeting summary, content research, email drafting, or vendor intake. One goal, few tools, clear output.
"Should the agent send emails automatically?"
Not for v0.1. Have it draft emails and send them to a human for approval. Add autonomy only after repeated successful tests.
§ 07
Plain-Language Definitions
Glossary
Use these definitions consistently across the workshop.
- Automation
- When something happens, a set of steps runs automatically.
- Trigger
- The event that starts a workflow, such as a form submission, email received, file uploaded, or row added.
- Input
- The data that enters the workflow, such as name, email, company, message, date, or file.
- Action
- A step the automation performs, such as adding a spreadsheet row, sending a notification, creating a task, or running an AI prompt.
- Output
- The result at the end of the workflow: saved record, message sent, draft created, decision prepared, or summary delivered.
- AI Step
- A workflow step where AI summarizes, classifies, extracts, drafts, or transforms information.
- Classification
- Assigning information to a category, such as urgent vs. not urgent or high-intent vs. low-intent lead.
- Human-in-the-Loop
- A review point where a human approves, edits, or rejects the system's output before consequential action.
- AI Agent
- A goal-driven AI system that can use tools and make decisions about how to complete a narrow task.
- Tool
- An external capability the agent can use, such as web search, email, Google Sheets, a CRM, or documents.
- Boundary
- A rule that limits what the agent can do, such as drafting but not sending emails.
- v0.1
- The smallest useful first version. It should work on one realistic case before you expand it.