PLB   Playbook

AI has raised the floor on design. Anyone with a chat window can now produce an ad, a landing page, or general marketing collateral that looks passable. By default AI designs better than most people do. The best designers, however, are still far ahead. AI copies well, picks up a theme, and can follow rules, but it struggles to make something original from nothing.

That leaves the gap between passable and great as the thing people notice. Stephen Haney, the founder of Paper.design, said in a Y Combinator interview that "if you look at every great company of the last 10, 20 years, basically all of them have exceptional design." Most of what we build rests on that belief. Put two products in front of us that work the same way, one designed well and one designed just fine, and we'll choose the better-designed one nine times out of ten.

Buyers judge the rest of us the same way. We recently worked with an engineering contractor on a project, then asked them for a proposal on a much larger scope, one which focused on front-end design and motion. What came back made it painfully obvious that our call notes had been fed into an AI tool and sent back without being checked or revised. The writing was generic and missed key context in several places. The layout was one we've seen hundreds of times, the version AI produces when someone asks for a report and gives it no guidance. A team pitching design work had shown that it didn't care how its own pitch looked, and that it was letting AI do the work without checking it for accuracy or quality. That told us what we needed to know, and they didn't win the work.

Every team we talk to is being asked to move faster. Bosses are demanding it and founders feel mounting pressure from the market every day. AI has turned that pressure into a story that slow teams won't survive. Design is a craft where this pressure is felt the most because good design takes intention, and creative work has no single right answer to race toward.

We've spent years around nonprofit marketing teams, and they show this challenge clearly. Many of them run like an internal agency for the rest of the organization. Fundraisers need decks, donor reports, and videos. Program teams need recruitment materials, event branding, and reports for partners, often with no more than a month's notice. A single event can call for an entire campaign that needs registration flows, reminder emails, printed materials, on-site collateral, and a follow-up plan. The colleagues asking for the work rarely see how long good creative takes, or how hard it is to carry that creative intentionally across every touchpoint in a campaign.

When communications comes in late, the work gets made piecemeal, each piece in isolation, and a program meant to feel like one brand ends up feeling improvised, fragmented, and ultimately short on quality.

Design rarely gets the runway it needs, and the quality of the product or program pays for it. AI won't hand that time back by itself. With a strong system behind it, though, AI gives a small team a fighting chance when a lot has to be made in a short window. The brand decisions are already encoded and first versions arrive in minutes, so the team's limited hours go to the questions only people can answer.

When we first brought AI into design work, we did what most teams do. We had brand guidelines, AI needed context, so we uploaded the PDF. What came back used our colors, and almost everything else was off. Each piece needed heavy cleanup from a designer before it was even passable.

The problem is who the guide was written for. A traditional brand guide is a document for human designers: agency partners, contractors, new hires. It gives them rules and trusts them to interpret, and good designers fill any gaps with judgment and taste. AI needs the same information encoded so it can read and apply it every time.

Harvey, the legal AI company, hit the same wall. In a post on how Harvey builds its marketing, co-founder Gabe Pereyra and creative director Shawn Farsai write that markdown files handled their colors and type fine, and that "markdown breaks one step up, when you try to explain how to make a chart for a blog post or structure a long-form PDF guide." Their answer was to rebuild their guidelines as a Figma library of templates and worked examples that agents can look at.

That's where design parts ways with writing. A voice guide can carry most of what makes writing sound like you. Much of what makes design look like you lives in proportion, spacing and placement, which are hard to put into words. In our content system issue, we said to write down your design system before asking AI to design anything. Writing it down turns out to be only part of the job.

The core is a token system: named values for every primitive in the brand, such as colors, fonts, type sizes and spacing, along with where each one is used. Around it sit written rules, an asset library of scalable vector art and imagery, and a set of components.

This is the skeleton we start from. Each section is encoded with tokens or design-tool variables, and we run it in Figma and Paper.

The section names will look familiar from any brand guide. What changes is what's inside each one: values an agent can read, rules for where each element is used and where it never is, and assets it can pull directly.

The rules are where taste gets translated. One brand we work with is built on a vibrant palette of many colors. How AI chose among them was the weak spot, so the system now spells out which colors to use, how often, what each one is never used for, and which pairings pass. Components work the same way. UX designers have always built with defined parts and rules for how they combine, and the approach carries over to brand and marketing design. Decide how a call to action looks, how many styles of it exist, what your headline treatments are and which motifs repeat, then encode each one.

Where it all lives depends on your tools. Our default is JSON token files plus markdown rule files, kept where any AI tool can reach them, such as a GitHub repo or a skill, with a design canvas like Figma or Paper on top. Agents build inside the canvas from the system, and designers can step in and adjust what they made. Keeping the system in a repo also means it moves with you to whatever AI tools you use next year. Teams on other stacks will set it up differently, but the parts matter more than the tooling.

Who sits at the keyboard matters too. A system run by designers working alongside AI can give the AI room to make layout choices. One run by marketers without design training needs tighter rules.

To start: don't ask AI to turn your PDF into a theme skill and stop there. Ask it to break the guide down into a token system and a set of markdown rule files, such as logo usage, font usage with web-safe fallbacks for email, and color. Then have it produce the collateral your team makes every month: an email, a set of ads, a flyer, a web banner, a landing page. The first round will lean toward your brand without landing. Critique each piece and turn every correction into a rule or a component. Expect long back-and-forth conversations at this stage, and expect the cleanup to shrink with each round.

Treat the first version of your design system as a minimum viable product. Ask what AI needs at minimum to produce work that comes close to your quality bar. For some brands that's colors and fonts. Others need vector art, or a photo library it can work with.

How much you need depends on how much brand equity you're protecting. A young brand without much history can start small. An established brand that people recognize by its look needs far more encoded before anything passes.

Once your system is established, grow it on purpose. The brands we work with refresh their creative every six months to a year, with new motifs or a shift in which colors lead, and each refresh is built into the system. Seasonal creative belongs there too, so holiday assets are ready to draw from the next time the season comes around.

Growth also comes from what goes wrong. Without strong grid and layout rules, AI arranges things logically and predictably, and the result looks like a machine made it. When that happens, write the rules it was missing.

With a strong system, AI gets 80 to 90% of the way on most of what we produce. We reserve at least 10% of a designer's time for every piece to take it the rest of the way. In our process, designers sit in three predictable places: the beginning, the middle, and the end.

At the start, someone with design expertise does the encoding. AI can build a design system on its own, but that tends to yield generic results. The person encoding needs a clear grasp of how the brand is identified visually, the strategy behind it, how color, photography, layout and type get used, and what the brand never does.

In the middle, on long builds like a UI system, designers check components and how the pieces come together before AI scales them.

At the end, designers add the final layer of judgment and polish. Much of that lives in details non-designers rarely notice: the space between letters and lines, the size of a logo against everything around it, how a headline sits against the body copy and the call to action. Handfuls of small details are most of the distance between passable and great. Motion belongs here for now. AI handles it decently and is improving fast, but the nuance of speed and physics in motion still requires clear input from people.

How much of that final pass a piece needs depends on its stakes. When we design a product, we run a skill that reviews the whole design, flags every behavior we haven't defined and finds the screens no designer has made yet, which are usually the error and edge states in a user journey. With strong tokens and components in place, AI can produce 20 or 30 of those screens in minutes, and a designer may never touch them. Their job is to inform clearly and stay on brand. Work meant to make someone feel something, or to sell, always gets the designer's final pass.

To start: sort what your team produces by stakes. Let AI own the informational pieces outright, and spend designer time on the pieces that sell.

One more shift is underway, mostly on the product side of our work. AI is closing the gap between designers and engineers. Designers on teams we work with now use agents to turn their own designs into front-end code, which removes the handoff where engineers guess at how hover and interactive states should behave. They still go back into a design tool when what they want can't be explained in words. We wrote about marketers becoming builders in an earlier issue, and design teams are on the same path. It leads designers to work more often alongside agents, directly in a codebase, and less often on a design canvas.

Ghost Brief builds design and marketing systems like this one alongside the brands we work with. If your team is trying to move faster without letting the work slip, we'd like to hear what you're building.