Stop Prompting, Start Shipping

AI Agents 101

Presented by Patrick McHeyser

Helping founders north of $1M ARR or funding grow without hiring by turning critical recurring work into AI-run systems.

The shift

From asking AI for an answer to putting AI inside a business process.

Two useful terms

Agent

An AI system that can pursue a task by deciding some of what to do next.

Agentic workflow

A repeatable business process where AI handles parts of the work inside a controlled structure.

The four pieces

What makes a workflow agentic?

Model

The reasoning engine. Claude, ChatGPT, Gemini, and the systems built on top of them.

Context

The information packaged for the model before it responds: question, product facts, docs, code, history, tone, and instructions.

Tools

External actions the workflow can call: search records, send messages, run checks, format text, or post back.

Goal

What done looks like. The standard the workflow can check against.

Case study

Customer success case study

Customer asks a product question in Skool
Founder or team has to notice it
Someone researches the answer
Careful response gets posted manually

In a larger Skool group, this was taking the team 5-10 hours a week depending on volume.

The goal

A response that is...

  • Accurate to the current product
  • Specific to the customer’s question
  • Clear and human-sounding
  • Careful about promises
  • Easy for a human to approve or edit quickly
The workflow
Customer question comes in A customer asks for help in Skool, which triggers the workflow.
Claude drafts with context Product facts, codebase, prior answers, tone, and instructions shape the response.
Humanizer checks the draft A tool catches stiff, generic, or obviously AI-written language.
Telegram approval Human review: approve, edit, or reject before it reaches the customer.
Approved response posts back A tool sends the approved answer back to the customer in Skool.
Mapped to the case study
Model Claude drafts and reasons through the customer question.
Context Customer question, product behavior, codebase, prior answers, tone, and instructions.
Tools Skool connector, Telegram approval, humanizer, and internal search.
Goal Accurate, specific, human-sounding, safe to approve quickly.
The takeaway
The model is only one piece. The workflow is where the leverage is.