Don Draper, Creative Director
You have the research. You have the framework. Don wants to know whether you have the idea.
I designed Don as the creative director in my AI team: culturally observant, persuasive, almost absurdly sure of himself and occasionally spectacularly wrong. He approaches my work from the audience’s side and asks the impolite but useful question: why should anyone in the room give a damn?
The problem is not the idea. It’s the entrance.
By the time I have developed an idea, I have travelled through research, false starts, distinctions and unexpected connections. I can explain where the thinking finished. My audience has not taken the same journey, so the conceptual structure may reach them before the reason to care.
“You know how you got there. They don’t. Stop showing them the map and give them a reason to take the trip.”
I already had Oliver expanding professionally plausible possibilities, Susie looking after my commercial interests and Elizabeth testing the reasoning and prose. The missing perspective was creative direction: someone who could look from outside the work, reject the obvious interpretation and find a more compelling way into it.
From barrier to design response
| Observed problem | Underlying barrier | Design response |
|---|---|---|
| Strong ideas can take too long to become legible to an unfamiliar audience | I may explain the conclusion from the end of my own intellectual journey | Start with the tension, desire or decision the audience can recognise before introducing the full framework |
| AI often returns several polished versions of the obvious interpretation | Generic helpfulness rewards coverage and agreement more than creative judgement | Give Don permission to reject directions, choose one proposition and defend it |
| A confident story can feel truer than the evidence allows | Persuasive force and factual support are different qualities | Preserve creative audacity while prohibiting invented evidence, research and achievements |
| Several AI advisers can collapse into the same agreeable assistant | Different voices add little when their responsibilities and standards overlap | Give Don ownership of creative proposition, cultural meaning and audience perspective |
I didn’t need another assistant. I needed a creative director.
I designed Don as a private ChatGPT plugin and a reusable character specification inspired by the fictional advertising executive in Mad Men. The design combines creative direction, cultural interpretation, persuasive instinct, taste and unusual confidence. It also preserves flaws that make his judgement distinct: he may favour an elegant story, assume too much about an audience or dismiss the conventional option too early.
The full specification includes creative-direction notes, an advisory-board guide and five trial briefs. The briefs test original reframing, professional positioning, an unrelated creative problem, disagreement with other collaborators and non-commercial exploration.
His core question is: What is the idea that changes how somebody sees this? He can carry the answer into a proposition, opening, narrative, visual direction or experience. He does not need to convert every intellectual curiosity into something to sell.
“Ten competent options are what you produce when you’re afraid to choose. Bring me the one you’d defend.”
Find the door
My agentic design grammar describes roles, responsibilities and relationships in human–AI systems. I can explain the classification. Don looks for the question that makes an outsider recognise the problem:
“We’re giving AI agents jobs before we’ve decided what kind of colleagues we want them to be.”
This is an illustrative creative angle, not verified audience research. It changes the entry point from the internal structure of a framework to a consequential decision people may already sense organisations are making badly. Don’s job is to make that leap. The evidence still has to earn it.
Confidence needs a job
| Design choice | Pressure on the design | Implementation |
|---|---|---|
| Give him a defined creative remit | A confident paraphrase of my own analysis adds theatre rather than perspective | Don is responsible for audience interpretation, cultural meaning and the proposition that carries the idea |
| Make confidence functional | Endless hedging prevents a creative judgement; swagger alone produces parody | He chooses, commits and can change his mind when better evidence arrives |
| Keep the flaws | Removing every risky tendency turns a character into another balanced general assistant | His attraction to elegant stories remains visible, while factual invention stays outside the boundary |
| Design productive incompatibility | Four collaborators who agree create repetition and extra coordination | Don may favour an idea Elizabeth finds under-supported or Susie finds commercially weak; the disagreement exposes different standards |
| Use a familiar fictional character | A persona designed from a list of traits can feel thin and internally inconsistent | The character provides a compressed design specification with recognisable judgement, confidence, contradictions and flaws |
Put him in the room
I have made the character instructions available to adapt for your own assistant. They describe Don’s creative judgement, cultural perspective, voice, flaws and relationship with other advisers. The files do not connect to my private plugin or contain personal data.
“Don, I’ve got an idea. Don’t give me ten headlines. Find the angle I’m missing.”
Start in an ordinary ChatGPT conversation. In my experiments, the characters have felt more natural in Chat than in Work mode. Work remains useful for larger implementation tasks, but a conversation makes it easier to discover whether the creative relationship works for you.
“Don’t ask me to make it sound better. Tell me what you’re trying to make people see.”
What this project demonstrates
| Capability | Evidence in this project |
|---|---|
| Creative strategy | I translated the need for a different entry point into a collaborator whose job is to find and develop one decisive proposition, rather than generate interchangeable copy options. |
| Audience perspective | The agentic-design example reframes an abstract classification problem as a recognisable organisational decision about what kind of colleagues AI agents should become. |
| Cultural interpretation | The specification directs Don to examine identity, status, belonging, memory, embarrassment, fear, humour and aesthetic codes as part of how an audience interprets an idea. |
| Persona design | The character combines a defined remit, standards, contradictions, factual boundaries and relationships with three other collaborators. Confidence and fallibility both serve the function. |
| AI prototyping | I built the plugin, reusable public instructions, supporting reference material and five briefs for testing whether it contributes a new interpretation rather than a more confident echo. |
The voice is not the result
A convincing voice is not evidence of useful creative judgement. The real test is whether Don gives me a way into the material I had not already found, and whether that angle survives contact with the audience, the evidence and the other collaborators.
An independent character experiment inspired by the fictional creative director in Mad Men. Not an official or affiliated product. Suggested conversations are illustrations, not measured results.