The Creative Unlock: Context, Taste, and the Act of Building

Last Friday I gave a lunch-and-learn to a performance-marketing agency I’ve worked with for years — the whole team on one call, sandwiches presumably off camera. Before I shared a single slide, I asked everyone to drop a number in the chat: rate your own AI ability, one to five.

Every answer came back a three. A 3.2 here, a 3.5 there. One of them — a builder who has stood up entire automation systems for other people — gave herself a 3.9, and then spent thirty seconds explaining why she didn’t deserve a four.

I’ve asked that question in a lot of rooms lately, and it’s always threes. Nobody feels like a five. And here’s what I’ve come to believe: the gap between feeling like a three and feeling like a five is not a tool problem and not a prompt problem. It’s three habits — the context you feed the machine, the taste you train it on, and the willingness to actually build. That’s the whole guide. Let me walk through the hour.

Nobody feels like a five — and it isn’t about the tools

This team is not behind. They have an agent stack wired into their Slack. One of them replaced search entirely — complex analytics problems go straight to Claude. Another built call-scoring rubrics that grade the team’s client calls against their own standards. The 3.9 has built AI systems for other companies.

And still: threes. Because when the conversation turned to creative work — ads, landing pages, design, video — everyone’s confidence dropped through the floor. “That’s a whole other skill,” the 3.9 said. “Prompting it accurately — that’s the part I’m trying to figure out.”

That sentence is where the entire hour pivoted, because it contains the misunderstanding almost everyone has.

Prompting is dead

When I asked the room how they’d close the gap on creative work, the answer was unanimous: find the right prompt. Get it perfect the first time so you don’t spend an afternoon going back and forth. One of them reached for the kindergarten exercise — tell the teacher how to make a peanut-butter sandwich, and if you’re not painfully specific she slaps the jar on the bread. If the machine gets it wrong, you weren’t specific enough.

I understand the instinct. It’s also the thing I most wanted to talk them out of. There is no magic prompt. The thought leaders offering to DM you theirs are selling you a souvenir. Even the labs building these models have moved on — the direction of the whole field is away from the clever one-shot instruction and toward structured context: giving the model the world it’s working in, then iterating inside a loop.

What that looks like in practice is folders, not phrasing. For my own brand I keep an operating file, a brand system, a design system, a foundation document — who I am, how I sound, what the work is for — pages of thinking that took real days to write. Every project gets the same treatment. When we built a campaign for one university client — fifteen landing pages, sixty ads — not one prompt was written until the folder held the story, the school, the visual identity, the images. The prompts afterward were almost boring. The thinking had already happened.

I’ve written a whole guide on the context half of this — The Company Is the Context — and another on why you should talk to your Claude Code instead of typing at it, because speech carries context your typing fingers edit out. So here I want to spend real time on the pillar nobody in that room had a system for.

Taste is trainable — and mood boards are the training

Here’s the uncomfortable question I asked them: you look at an AI-generated ad and say that’s garbage — fine, your eye works. But have you taught the machine what your eye knows? Almost nobody has. We expect the model to have our taste without ever showing it our taste, then call the output slop.

The fix is embarrassingly old-fashioned: mood boards. The exercise every creative director has dragged you into, the one that makes analytical people roll their eyes. It has quietly become the most valuable thing I do.

My version is a running visual database that Claude Code maintains for me — an HTML gallery of styles, references, reactions. A year ago I had never heard of Memphis design. Or neo-brutalism. Or vaporwave. Now they’re vocabulary: I can say give me this in Memphis style and the machine and I mean the same thing. That’s what a mood board actually is — not decoration, a shared language.

Where does the material come from? Everywhere you already look. An ad in the wild that stops my thumb — screenshot, into the board, “make me eight versions of this energy with our context.” A website whose scroll feels alive — into the board. Museums: I’ve photographed paintings my whole life and texted them to designers who, reasonably, had eighteen other things going on. Now I text them to my terminal and the answer comes back the same afternoon. The big New York museums have put centuries of design online; every great graphic designer you admire is already studying it. That’s how influence has always worked — you collect what moves you and make it yours. The machine just collapsed the distance between seeing something and learning from it.

And then — this is the part people skip — you iterate past the point of comfort. My homepage took roughly three hundred versions. Different boards, different styles, different failures, until one appeared that felt like mine. The illustrated figure on my site took eighteen attempts before I accepted it as me — I needed to see the likeness from enough angles to trust it as a scalable style, not a lucky frame. When I show people the first landing page the machine made me, with no context and no board, they laugh. It looks like everyone’s landing page. That’s what no-taste output is: the average of the internet.

It sounds like a ton of work. It is. Once. Then you own a system — a trained eye the machine can borrow — and every project after that starts from your taste instead of from zero.

The act of building — the pipeline is shorter than you think

The third pillar is the one that scares marketing people, and it shouldn’t. The entire pipeline I use to ship real, client-facing work is four moves:

Idea → Claude Code → GitHub → Netlify.

Claude Code writes the code while you talk to it in plain English. GitHub keeps the versions. Netlify makes it a live URL. That’s it — that’s the machinery behind every landing page I send a client. The quiet superpower is that the link never changes while everything behind it does: I can keep improving a page for weeks and the client just sees the same URL getting better.

At this point in the hour I stopped presenting and made everyone open their terminal. Most of them had never seen it. One line to install, type claude, and there’s the alien.

The whole on-ramp — one line in the terminal, one word, and then you’re just talking. Nobody in the room had to read a line of code.

Two people hit command not found. One of them did the single most human-in-the-loop thing I’ve ever watched: she asked her Claude why the install wasn’t working, pasted in what it told her, and thirty seconds later — “now it’s working.” She debugged a developer tool without being a developer, by asking. That’s the entire lesson of the third pillar in one move. You don’t have to read the code. You don’t have to understand the code. You have to be willing to stay in the conversation.

Because making a video is coding now. Making a design is coding. The deck I presented from — the slides, the animation, the little lights-on opening — was built the night before in one Claude Code conversation. Twenty-one years into my career, the scariest tool on my computer turned out to be the most creative one.

Where to start on Monday

If the three pillars feel like a lot, here’s the honest on-ramp — the same one I gave the team:

1. Context: write one operating file. Pick one client or project. Open a folder, write a page that says what this thing is, who it’s for, what it sounds like, and what good looks like. Don’t prompt anything until that page exists. (The full system is in The Company Is the Context.)
2. Taste: start the board. Make a folder called moodboard. For one week, every ad, site, painting, or poster that stops you goes in with one sentence about why. At the end of the week you’ll have a design brief no agency could have written for you, because it’s your eye on paper.
3. Building: install the thing. Open the terminal, run the one-line install, type claude, and describe a real problem in plain English. When it breaks — and something will — ask it why. That’s the whole skill.

None of these takes a budget. All three take the one thing the machine can’t supply: your attention.

A team of ones

I’ll say the honest thing I said to them at the end. I used to work alongside dozens of people to produce what I now produce with a small crew and a terminal — and I don’t find that scary, because the point was never subtraction. When one person can carry an idea from folder to mood board to live URL, everyone on a team becomes capable of being their own team. The companies that get this won’t shrink; they’ll compound.

But none of it comes from a prompt. It comes from the thinking you structure (context), the eye you train (taste), and the conversation you refuse to quit (building). Three habits. The threes in that chat window are closer to five than they think — and so, probably, are you.

— Joe