Two people, eight teams, one locked door
At MoonPay, one craft was locked behind two people: the words in the product. Content design. Two content designers were covering more than eight product teams, and that coverage never stretches far enough, so they ended up guarding the standard instead of spreading it. Underneath was a lack of trust: they didn't believe the rest of the org could write well enough, so the surest way to hold the bar was to keep the work themselves. That is a gate, and every org has one like it, some craft only a couple of trusted people are allowed to do. The real cost isn't the bottleneck. It's that everyone else quietly decides the craft isn't their job. Twenty percent of the org got a hundred percent of the way there; the rest were on their own, and had quietly learned that words just weren't theirs. So I set out to kill the model that kept it locked.
Lorem ipsum wasn't a placeholder. It was a decision nobody had made.
In a critique, "ignore the copy, it's still lorem ipsum" passes without comment. It shouldn't. That phrase assumes the screen is basically finished and the words drop in at the end. It's backwards. The words are what the screen is for: they decide what leads, what gets cut, what even needs to be there. Lorem ipsum isn't a placeholder, it's a decision nobody has made yet, on its way to production looking finished. A designer who hands the words off is doing half the job. Words are seasoning: worked in throughout, they become the flavour; thrown on at the end, they just sit on top, and the screen can look perfect and still read wrong.
You can't hire your way out of a gate
Hiring more content designers was the obvious fix, and also the trap. We were interviewing for two, and the candidates were good, but more specialists is more gate: it says, louder than before, that the words are their job and not yours, and it doubles down on the model that got us here. The math never allowed it anyway. Demand grows with every team and every surface; hiring grows one head at a time, so the gap only widens. The problem was never how many specialists we had. It was that only specialists were allowed to write.
You can't ask someone to dismantle their own gate
Before building anything, I tried to get the content team to lead this themselves. It went nowhere. I'd asked them to show me how we scale past throwing bodies at the problem, and nothing came back. For a while I read it as resistance. It wasn't. I was asking them to dismantle the thing that made them feel safe: if everyone can write, what's the part that's still just yours? That's not politics, it's the fear of being made replaceable. You can't hand a change to the people it frightens, and you can't ask anyone to volunteer for their own redundancy. The leader carries that risk, and proves on the other side that it makes them more valuable, not less. So I picked it up myself.
I didn't change a single rule. I built a way to deliver them.
A custom GPT was the cheapest possible test: no engineering, no roadmap slot, no budget. If I was wrong, I'd lose a weekend, not a quarter. The question it answered was simple: put the team's standards where people actually work, and can anyone hit the bar? And I changed nothing about how we write. The team had spent eighteen months building that: voice, tone, reading levels, glossaries for how we name things internally versus to customers, how errors get handled. A huge repository of genuinely good work, already there. My job was never to improve the rules. It was to make them reachable.
How we write was only half the gate
The rules were the easy half. The other half was the knowledge you need to use them in a regulated product:
- What the FCA allows, the line between a fair claim and a regulated promise.
- How our product actually works, which lived in our FAQs.
- How the best fintech writes, the patterns the strongest products already use.
Normally a designer would chase all of that down themselves, one Slack thread and one legal review at a time. So I wired the writing rules straight to those three sources. Nothing exotic, just the rules sitting next to the knowledge that makes them usable, the moment you start.
"Audit this screen."
I tested it on a screen I didn't trust: our live MoonPay Balance modal. The prompt was three words. Audit this screen.
- 1ClaritySells what it is, not why you'd want it.
- 2ComplianceAn unvalidated promise, already shipped: “near perfect approval rates.”
- 3DensityFive bullets where people read three. The last isn’t a benefit.
- 4CTA“Let’s go” never says what happens next.
The live MoonPay Balance modal I handed to the tool, the one I didn't trust.
No steering, so all it had was the rules and context I'd built in. It came back with a strategist's read, and most of it was craft: the screen sold what the feature was, not why you'd want it; five bullets where people read three; a "Let's go" button that said nothing about what came next. But one finding stopped me. It flagged "near perfect approval rates" as a claim that needed compliance sign-off, a regulated promise that was already live. It wasn't reasoning about spelling. It was reasoning about clarity, conversion, and a compliance risk nobody had raised. That was when I knew this was more than a writing aid.
One unblocked person at a time
Adoption didn't come from a rollout. It came from the people who were drowning. I never announced it. I handed it quietly to the designers with no content support, and they grabbed it because they needed it. Once they had it, they stopped waiting for someone else to own the words and started owning them. The "it's still lorem ipsum" excuse disappeared, because the work arrived already good. One unblocked person at a time, until everyone who shipped owned the words, not just the two specialists.
It changed the conversation with legal
The clearest proof showed up somewhere I hadn't planned: the legal reviews. The hardest part of the job is the legally sensitive copy, turning what compliance allows into something a customer can actually understand. That used to get handed up to a content designer or a PM. Now the solo designers could do it themselves, and they walked into legal reviews already fluent in the constraints. The relay that used to carry legal's notes back to design, round after round, mostly disappeared. It never replaced legal; legal still made the call. But designers were suddenly doing cross-functional work they'd never have touched, and the collaboration got better for it.
It made the design better, not just the words
Using it daily showed me the ceiling: a screenshot is an island. The tool could read one static image, but it knew nothing about the design system behind it or the rest of the file. So I built it into Figma as a plugin, where it could read the real structure, draw on the system, and see a screen in the context of the whole flow. That changed what it did. Once it could see the structure, it stopped suggesting better words and started suggesting better structure: move this up, this should lead, you don't need this at all. The design started to follow the content, instead of the words getting bolted onto a finished layout. It didn't just help people write better. It helped them design better.
What it added up to
The old model leaked three ways, and the tool closed all three.
- Consistency. Two of eight teams became every team, on the same bar.
- Cost. Problems that used to surface after launch now surface at the keyboard, the cheap end of the curve.
- Speed. No one waits on a content pass.
It spread org-wide with no rollout email, which is the only signal I really trust. The £650,000 a year it saved in headcount was never the point. It's the receipt.
From scraping under the bar to building the ceiling
Most of the old content job was defensive: wording something just enough to clear legal and the FCA. That's scraping under the bar, not raising it. Once the tool carried those constraints and the repeatable work was something anyone could do safely, the specialists came off the grunt work for good. They stopped negotiating the floor and started building the ceiling. In a crypto product, the ceiling is the words that carry feeling:
- The line that calms someone watching a transaction sit pending.
- The voice that makes creating a wallet feel like the frontier, not a liability.
- The honesty when the network fails and it's out of our hands.
That isn't decoration. It's the difference between a product people use and one they trust. The bar stopped being who does crypto well. It became who makes money feel human.
From doing the work to authoring it
The specialists changed jobs without changing seats. They stopped brute-forcing strings team by team and became the people who define how the whole org writes. That's the real lesson: you don't scale a craft by doing more of it yourself, you scale it by democratising access to it. Do that to every craft and people start building across the lines that used to separate them. Nobody stops being the expert; each craft still owns its piece. What's new is the middle, where the lines blur, and what sits there is a builder, one person who can carry an idea from start to finish. I'm not predicting that. I've been living in the early version of it.
An expert's job isn't to do the work
For years I'd have told you an expert's job is to do the work, to be the best in the room at the thing. I don't believe that anymore. An expert's job isn't to do the work. It's to make everyone else as good at it as you are, or better.
