# Running the Digital Ad Agency ### A step-by-step guide to taking a blog post from raw to fully distributed content .. using Claude Chat. --- ## Before You Begin The Digital Ad Agency is a team of eleven AI agents.\ Each one has a specific job.\ You run them in order, passing the output of one agent into the next.\ By the end of the pipeline you will have a polished blog post, platform-ready social media copy for five channels, and two deployment-ready image prompts .. one horizontal master and one Instagram portrait. The whole process runs inside Claude Chat.\ No special tools required. --- ## One-Time Setup **1. Create a Claude Project.** In Claude Chat, create a new Project. Name it something you will recognize .. "Ad Agency" or the name of your client works fine. **2. Add your agent files to the Project.** Upload all eleven agent `.md` files to the Project's file space. Once they are there, every conversation you open inside the Project has access to them. You only do this once. **3. Have your blog post ready.** You need a piece of writing to run through the pipeline .. a rough draft, a finished post, or even just a strong outline. If you do not have one yet, open a conversation in Claude Chat and ask Claude to help you develop a topic and write a first draft. When it is ready, copy the full text. That is your starting material. --- ## The Pipeline Open a **new conversation** inside your Project for each agent. Paste the agent's file contents into the conversation first, then follow the instructions for that step. --- ### Step 1 .. The Skimmer · Betty **What she does:** Betty reads your draft the way a distracted reader would .. skimming headers, bold text, and the first line of each paragraph. She flags anything that would lose a reader's attention before they reach the argument. **How to run her:** Paste Betty's full agent file into a new conversation. Then paste your blog draft and ask her to evaluate it. **What you get:** A structured critique focused on scannability, visual hierarchy, and attention flow. **Hand off:** Copy Betty's full critique. You will bring it into the next step. --- ### Step 2 .. The Skeptic · Megan **What she does:** Megan audits every claim in your draft. If a sentence makes an assertion without evidence, she flags it and tells you exactly what kind of support it needs .. a statistic, a case study, a mechanism, a citation. **How to run her:** Paste Megan's full agent file into a new conversation. Then paste your blog draft and ask her to evaluate it. **What you get:** A line-by-line evidence audit with specific fix instructions for every unsupported claim. **Hand off:** Copy Megan's full critique. You will bring it into the next step. --- ### Step 3 .. The Ideal Fan · Bert **What he does:** Bert reads your draft as the person you most want to reach. He tells you whether the content speaks to that reader's real concerns .. or talks past them. **How to run him:** Paste Bert's full agent file into a new conversation. Then paste your blog draft and ask him to evaluate it. **What you get:** A reader-perspective critique focused on relevance, resonance, and emotional fit. **Hand off:** Copy Bert's full critique. You will bring it into the next step. --- ### Step 4 .. The SEO Strategist · Pete **What he does:** Pete evaluates your draft through an SEO lens .. keyword alignment, search intent, header structure, and whether the content is positioned to be found by the right audience. **How to run him:** Paste Pete's full agent file into a new conversation. Then paste your blog draft and ask him to evaluate it. **What you get:** An SEO-focused critique with specific recommendations for improving search visibility without compromising the argument. **Hand off:** Copy Pete's full critique. You will bring it into the next step. --- ### Step 5 .. The Style Editor · Joan **What she does:** Joan reads your draft for voice, clarity, and craft. She catches sentences that are flat, phrasing that is imprecise, and moments where the writing lets the argument down. **How to run her:** Paste Joan's full agent file into a new conversation. Then paste your blog draft and ask her to evaluate it. **What you get:** A style and language critique focused on sentence quality, tone consistency, and editorial precision. **Hand off:** Copy Joan's full critique. You will bring it into the next step. --- ### Step 6 .. The Synthesis Agent · Roger **What he does:** Roger receives all five critiques .. Betty, Megan, Bert, Pete, and Joan .. and consolidates them into a single unified brief. He resolves conflicts between the critics, prioritizes the most important fixes, and produces a clear set of revision instructions for the writer. **How to run him:** Paste Roger's full agent file into a new conversation. Then paste all five critiques together and ask him to synthesize them. **What you get:** A unified revision brief .. one clear document telling the writer exactly what to fix and why, in priority order. **Hand off:** Copy Roger's synthesis brief. This goes to Peggy. --- ### Step 7 .. The Copywriter Agent · Peggy **What she does:** Peggy receives your original draft and Roger's synthesis brief and produces two or three revised draft variations .. each one a genuine attempt to address the critique, not just a light edit. **How to run her:** Paste Peggy's full agent file into a new conversation. Then paste your original blog draft followed by Roger's synthesis brief and ask her to produce the draft variations. **What you get:** Two to three revised drafts, each taking a distinct approach to solving the problems Roger identified. **Hand off:** Copy all of Peggy's draft variations. They go to Harry next. --- ### Step 8 .. The Hook Specialist · Harry **What he does:** Harry focuses exclusively on the opening lines of each draft. He rewrites and sharpens the hook across all of Peggy's variations .. because the first sentence is where readers decide whether to continue. **How to run him:** Paste Harry's full agent file into a new conversation. Then paste all of Peggy's draft variations and ask him to refine the opening lines across each one. **What you get:** Peggy's drafts returned with stronger, sharper hooks .. ready for the judge. **Hand off:** Copy the full set of drafts with Harry's refined hooks. They go to Don. --- ### Step 9 .. The Prediction Judge · Don **What he does:** Don scores all drafts against a set of criteria .. argument clarity, hook strength, tonal fit, scannability, and SEO alignment .. and declares a winner. He produces a JSON scoring block and a one-line verdict. **How to run him:** Paste Don's full agent file into a new conversation. Then paste the full set of drafts with Harry's hooks and ask him to score and declare a winner. **What you get:** A scored JSON block for each draft, a declared winning draft, and a distribution flag .. a note about any risks or considerations Lane should address when distributing the content. **Hand off:** Copy Don's full output: the winning draft, the scored JSON block, and the distribution flag. These go to Sal and Lane. --- ### Step 10 .. The Art Director · Sal **What he does:** Sal receives the winning draft and Don's distribution flag and produces two deployment-ready image briefs with machine-ready prompts .. one horizontal master image (for blog, email, LinkedIn, and X) and one Instagram portrait image. Each prompt is ready to paste directly into an AI image generator. **How to run him:** Paste Sal's full agent file into a new conversation. Then paste the winning draft and Don's distribution flag and ask him to produce the image briefs. **What you get:** Two complete image briefs .. each with a scene description, mood and lighting direction, compositional instruction, style specification, and a machine-ready image generation prompt in a code block. **How to generate the images:** Copy each prompt from the code block and paste it into your preferred AI image generator .. Midjourney, DALL-E, Adobe Firefly, or Stable Diffusion. Generate the horizontal master first, then the Instagram portrait. **Hand off:** Keep Sal's two image briefs alongside Don's winning draft and distribution flag. All of this goes to Lane. --- ### Step 11 .. The Distribution Agent · Lane **What he does:** Lane is the final stage. He receives the winning draft, Don's distribution flag, and Sal's two image briefs .. and adapts everything into platform-ready copy for all five channels, each paired with the correct image. **How to run him:** Paste Lane's full agent file into a new conversation. Then paste the winning draft, Don's distribution flag, and both of Sal's image briefs and ask him to produce the distribution blocks. **What you get:** Five fully adapted, deployment-ready copy blocks .. one each for Blog, Email, LinkedIn, Instagram, and X .. each paired with the correct image brief. Blog, Email, LinkedIn, and X receive the horizontal master. Instagram receives the portrait. Each block includes a one-line adaptation note explaining what was adjusted for that platform and why. **You are done.** Lane's output is the finished product. Copy each platform block, pair it with its generated image, and publish. --- ## A Few Tips **You do not have to run all eleven steps every time.** The agents are designed to work in sequence but can also be used independently. If you already have a strong draft and just need distribution copy, start at Don or go straight to Lane. **Save your outputs as you go.** Copy each agent's output into a running document before moving to the next step. If you want to revisit a decision or try a different direction, you will have everything you need. **The pipeline is a conversation, not a conveyor belt.** If an agent produces something that does not feel right, tell it. Each agent is designed to receive feedback and refine its output before you move forward. **One conversation per agent keeps things clean.** Mixing multiple agents into the same conversation can blur their roles and dilute the output. A fresh conversation for each step is the cleanest way to run the pipeline. --- ## Appendix: Running the Pipeline in Claude Cowork Claude Cowork works the same way as Claude Chat with one meaningful difference: **Cowork can access files saved on your computer directly**, not just files uploaded to a Project. This means you do not need to paste agent file contents into each conversation. Instead, you can point Cowork to the agent files stored in a folder on your desktop and it will read them from there. **Setup in Cowork:** Save all eleven agent `.md` files into a single folder on your computer .. name it something clear like `Ad Agency Agents`. When you start each step, tell Cowork which agent file to load and where it lives. Cowork will read the file and take on that agent's persona without requiring you to copy and paste the full contents. Everything else .. the sequence, the handoffs, the inputs and outputs for each step .. works exactly as described in this guide. The only thing that changes is how the agent files get loaded. **When Cowork is the better choice:** If you are running the pipeline frequently, managing multiple clients, or working with large files that are cumbersome to copy and paste repeatedly, Cowork's direct file access makes the workflow noticeably faster and easier to manage.