Building the Anon. ecosystem with Claude

Designing early concepts of a coherent digital brand across website, magazine and talent portal.

Two prototypes, built to test how the same brand language holds up across very different jobs — a public-facing showcase and a twice-yearly editorial title.

An AI Persona Built to Test Design Judgment

Designing a persona-driven learning tool that tests judgment, not just execution — built, debugged, and refined through real testing.

The execution layer junior designers once cut their teeth on is disappearing. AI and automation are absorbing the roll-out work — typeface exploration, first-pass layouts, production polish — that used to be how a junior earned their stripes. What's left, and what's now more valuable than it's ever been, is design thinking: the ability to interpret a brief, make a defensible decision, and communicate that decision clearly.

It's no longer enough to design well. It's crucial that a designer can think on their feet in the room — knowing when to take feedback on board and when to hold their ground, and being able to say why, clearly and quickly, to someone who isn't a designer at all.

Foundry is a platform built around that gap. Real design projects, worked without a safety net, presented once to an AI client with a genuine, consistent point of view — who pushes back, asks hard questions, and has to be persuaded on their own terms. This project is the AI persona at the centre of it: how it was built, what broke, and the decisions that shaped it into something that actually teaches that skill rather than just performing it.

The problem

Design school and peer crits teach you to defend your work to people who think like designers. Almost nobody teaches you to defend it to someone who doesn't — which is what actually happens the first time you present to a real client, and by then the stakes are real.

The mechanic

Four monthly projects, each across brand, art direction, graphic design, and UI/UX — and each with its own client persona to navigate. Work it your own way, in any order, with your reasoning documented as you go. Present the finished thing once, in a single session with no safety net, to that project's AI client — a real, consistent point of view who pushes back, asks hard questions, and has to be persuaded on their own terms. Then get a written crit from a design director on the work itself — because a client's approval and good design aren't always the same thing, and you need to know the difference.

Building the persona — a personality wasn't enough

Early versions of Marcus Hale, a founder building a pet insurance brand, had personality traits — blunt, impatient — but no fixed point of view. Every response was improvised fresh, so feedback contradicted itself session to session: the same logo praised in one run, rejected as generic in the next, with nothing stable for the model to reason against.

The fix was a grounded taste profile, not just a character sketch: a stated core anxiety (this business survives on getting noticed, or it dies invisible), concrete likes and dislikes specific enough to check work against, and an explicit rule for what actually earns a reversal — arguments tied to attention and distinction, not confidence or design vocabulary alone.

What broke, and what fixed it

Same logo, shown in two separate sessions. First verdict: "the pet heads in the marque worked well." Second verdict, same image: "generic — looks like every other pet logo." No memory, no ground truth to check against — the model was improvising a plausible opinion each time, and even contradicting its own stated rules.

The fix was a decision log that records not just what was approved, but why — and a rule that an unargued position can't reverse for free. Only a reason tied to the persona's core anxiety earns a change of mind, and once earned, that new position holds.

A deliberate constraint — why there's no second session

An early version allowed two sessions per brief. Testing surfaced the real risk directly: if the client can be swayed mid-project, a designer could spend real hours rebuilding around one imperfect opinion, chasing a direction that was never guaranteed to be right. The fix wasn't more coaching — it was one session, full stop.

"Share your final version. You get one session. Make it count."

Two personas, two pressures

Every project's persona is built to test something different, not just to vary the flavour. Marcus Hale — a startup founder, presented over a Zoom call — is blunt and confrontational; defending work to him means holding a position under direct, adversarial pressure. Priya Ashworth — a bakery founder, presented as an in-person walkthrough — is warm and indirect, and her real objections hide behind "it's lovely, but..." Same underlying mechanic, deliberately different psychology, because a designer who's only ever practised handling one type of client hasn't really practised the skill at all.

Sharpening the interrogation

A weak version of the persona only reacted to what was directly in front of it. The rebuilt version asks specific, adversarial-but-plausible follow-up questions, adapted to what's actually shown — the kind of thing a real, moderately tech-literate stakeholder would ask:

What a designer walks away with

A real, unaided decision, made with no mid-project feedback to lean on. Practice defending it under specific pressure from someone who isn't a designer. The distinction — hard-won, not just told — between winning an argument and having made good work. Specific, named craft feedback to revise against. And, at the end of it, a stronger portfolio piece — built across four different projects, four different personas, four different kinds of pressure.