This Is What It Looks Like to Build a Real Agent in Claude Cowork

Most people open Claude, type a request, and get something generic back that needs cleanup. That's not a Claude problem, it's the difference between chatting with AI and actually running an agent. This video shows how to build a real AI agent inside Claude Cowork, no code required, using a real viewer's job as the example: reviewing behavior data, drafting improvement plans, and writing a weekly report. You'll see the exact file that makes an agent work, the three-phase workflow design process, and how to schedule it to run on its own. __________________________________________________ BEFORE READING MORE — were you aware of this? Every time you switch tools, change jobs, or get handed a new AI at work, you start over from scratch. That context stays behind. The version of AI that actually knew you? Gone. You spend the next few weeks re-explaining yourself, getting outputs that are close but off, wondering why it stopped working the way it used to. GET YOUR BEST CLAUDE.md FILE HERE: If this is for work that actually matters for your business or your career, here's how to fix it. The Professional AI Interview Kit: https://calhyslop.com/now-what Before it asks you a single question, it reads whatever context already exists in your workspace. Then it interviews you across ten areas, including your professional history, the things you've carried from past roles that shape how you work today and would never come up if someone just asked what your job title is. The result is a Personal Operating Document that reflects what you said and what it found. Yours to keep. Yours to carry. Paste it into any tool, any time. The AI stops guessing. Includes a step-by-step video walkthrough. Setup takes under five minutes. There's also a free version built for personal use. Not as complete, but a real starting point if this is just for you, not your job. The Claude Cowork Interview Prompt (Free): https://calhyslop.kit.com/the_claude_cowork_interview_prompt Instead of filling out a document on your own, this walks Claude through interviewing you directly. Just paste it in and Claude asks the questions, one at a time, then builds your CLAUDE.md file from your answers as you go. The gap worth knowing about: Claude can only ask what it thinks to ask. The patterns you carry from past roles, the standards you hold without naming them, the professional context that shapes everything you do but never comes up when someone asks what your job title is, those go unasked unless something prompts the question. Which means the AI is still guessing about the things that matter most. The kit is specifically built to surface exactly those things. __________________________________________________ Tools I Use Every Day (And You Probably Should Too) Typeless — Stop typing. This replaces your keyboard. It converts your speech into clean, formatted text in any app with two keystrokes. No switching apps, no recording interface. Just press, talk, press. If you've ever lost a thought mid-sentence because you were too busy typing to finish it, this is the fix. Desktop and mobile. The link below includes a $5 credit when you sign up. 🔗 typeless.com/?via=cal (https://www.typeless.com/?via=cal) | Invite code: cal __________________________________________________ What You Will Learn ✓ The Three Levels of Working with AI — Understand the real difference between basic chat, builder mode, and running a full agent, and why most people stall out one level below where the payoff starts. ✓ The CLAUDE.md File — Learn what this file is, why Claude reads it automatically every session, and the four things it needs at minimum to actually work. ✓ Designing a Multi-Phase Agent Workflow — See how to describe an entire job in plain English so Claude builds a reusable plan instead of guessing at each step. ✓ Turning a Plan into a Reusable File — Watch a workflow plan get saved as an actual file the agent can run again without you repeating anything. ✓ Running the Agent on Real Data — Watch the agent carry out all three phases on its own: finding patterns, drafting improvement plans, and writing a weekly report. ✓ Scheduling the Agent to Run Itself — Learn how to set the workflow to run automatically instead of triggering it by hand every week. ✓ The Mistake That Kills Most Agent Builds — Understand why skipping the plan review before approval is what turns an agent into a guessing machine. 🕒 TIMESTAMPS: 00:00 The Structural Problem with AI Chat 01:14 A Real-World Use Case 03:17 The 3 Levels of Working with AI 07:37 Step 1: Design the Workflow in Plain English 09:31 Step 2: Creating the Reusable Agent File 10:08 Step 3: Running the Agent End-to-End ABOUT CAL With over 20 years of experience as an educator and a deep background in AI automation, Cal Hyslop helps experienced professionals build AI workflows that actually fit their work — without writing code or becoming an AI expert. #AIAutomation #PracticalAI #AIForProfessionals #AIAgents