Build Your First AI Agent Workflow

What will I be able to do after this training?

• Build a multi-step agent workflow inside Claude Cowork that plans, executes, and learns from feedback
• Create a customer service automation that pulls emails, searches a knowledge base, drafts replies, and sends after your approval
• Turn wireframes + transcripts into website copy using the Draper brand voice and Growth OS context files
• Feed a Story Brand brandscript into Claude Cowork and generate a social media content calendar
• Follow the "plan first, then implement" pattern that makes Claude Cowork workflows reliable and repeatable
• Turn any completed workflow into a reusable skill so Claude can repeat it without re-prompting

What key terms do I need to know?

• Agent: an AI that takes multiple actions across long time horizons, makes plans, connects to external tools, and remembers instructions. Not just a chatbot.
• Skill: an instruction file (skill.md) that Claude creates and self-edits inside your folder. It tells Claude how to perform a specific task. You don't write it. Claude does.
• Workflow: a multi-step process that chains multiple skills and connections together to accomplish a larger goal (e.g., pull email, search KB, draft reply, send)
• Growth OS: the folder structure and plugin system that organizes your business context (core files, workflows, routines) so Claude can work effectively. See the Claude Cowork for Marketers training for a full introduction.
• Quick Start: a plugin that sets up your Growth OS folder in about 10 minutes, including persona, offers, business info, and initial skills
• Routine: a scheduled sequence of skills that runs on a trigger, like saying "good morning" to kick off a daily workflow
• External Connections: integrations with tools like Gmail, Zoom, Notion, Slack, Google Calendar, and Beehiiv that let Claude act in the real world
• Story Brand: Donald Miller's marketing framework based on the hero's journey (character, problem, guide, plan, call to action, success, failure)
• Draper: a brand voice style (named after Don Draper) that writes punchy, persuasive marketing copy. Claude can auto-detect when to use it.

Demo 1: Customer Service Rep

Built by Russ

What it does: Pulls support emails from Gmail, searches a knowledge base for answers, drafts editable replies, and sends after approval.

How it was built:

1. Described the end result: "Pull support emails, search a knowledge base, draft a reply, let me approve, then send."
2. Claude asked which Gmail folder to pull from. Answer: the support label.
3. Claude created the skill file, the knowledge base files (membership/billing, product/tech support, trainings/events, general inquiries), and wired up the Gmail connection.
4. Tested with a real email. Claude auto-skipped spam, identified the pricing question, found the answer in the knowledge base, and drafted a reply.
5. Requested editable file links instead of prompting back and forth. Claude updated the skill to support that.
6. Added to the morning routine so it runs every day automatically.

Key insight: When something was wrong (no greeting, too direct), just explain it. Claude edits the skill and fixes it for next time. The skill self-heals.

Demo 2: Website Copy from Wireframe

Built by Julie (4 AM from Australia)

What it does: Takes a wireframe PDF and writes complete website copy, then builds the full HTML page.

How it was built:

1. Set up Growth OS Quick Start for the client's business (motorcycle training academy).
2. Fed the wireframe PDF and said: "Analyze this wireframe, list the copy requirements."
3. Claude returned a structured list: hero headline, subhead, trust badge, video block, body paragraphs, mission section, etc.
4. Context was already in the folder (persona, offers, brand voice, event transcripts from the Quick Start setup).
5. Claude auto-detected the Draper brand voice and applied it without being asked.
6. Said "write the copy" and Claude produced the full page copy in one turn.
7. Then said "build the HTML page" and got a working page in ~20 minutes total.

Key insight: The page compared favorably to the developer-built version that cost 10K+. Context from Growth OS (persona, offers, transcripts) made the copy specific and accurate, not generic.

Demo 3: Story Brand Social Media Calendar

Built by Fred

What it does: Takes a Story Brand brandscript and generates a structured social media content calendar for Meta (Facebook/Instagram).

How it was built:

1. Had an existing Story Brand brandscript built in ChatGPT for Miles Auto Service (auto repair shop) with three customer personas.
2. Fed the brandscript file into Claude Cowork's Growth OS folder.
3. Prompted: "Follow this file to make Instagram posts. Create a plan to implement this in my folder."
4. Claude asked clarifying questions: weekly or monthly? How many posts? What tone?
5. Built the calendar in a workflows/meta-calendar folder automatically.
6. Final step: "Analyze everything you did and build a skill that builds this calendar every time."

Key insight: Existing ChatGPT assets (custom GPT instructions, project docs) transfer directly into Claude Cowork. Copy the instructions as Markdown and tell Claude to build it as a skill.

How do I get started?

1. Install the Quick Start plugin. Go to theclick.ai/training and follow the Claude Cowork Quick Start training. It installs a Growth OS folder with core files, workflows, and routines in about 10 minutes.
2. Watch what Claude builds. Keep your folder open while working. Watch how Claude organizes files. If you don't like the structure, tell it.
3. Start with one workflow. Pick a task you do regularly (email replies, social posts, content creation) and build it as a workflow using the pattern above.
4. Build the skill. Once the workflow works, have Claude turn it into a skill. Now it's automated.
5. Connect external tools. Add Gmail, Zoom, Notion, or other integrations to let your workflows interact with the real world.

Pro tips from the session

• Copy as Markdown. When moving documents from ChatGPT or anywhere else into Claude Cowork, copy as Markdown for the best results.
• Don't over-specify the how. Tell Claude the end result. Let it figure out the path. These models are far more capable than the chatbot era.
• Give examples when correcting. Saying "that's too hypey" is good. Saying "that's too hypey, for example, this sentence right here" is much better.
• Migrate your GPTs. Copy the instructions from any custom GPT you love, paste them into Claude Cowork, and say "build this as a skill." Done.
• Use the biggest model. For complex, multi-step workflows (wireframe to copy to HTML in one turn), use the Opus model for best results.
• Context is everything. The more business context Claude has (persona, offers, transcripts, brand voice), the better the output. Growth OS builds this foundation. Use AI Deep Research to feed even richer data into your workflows.