Your clickable prototype is already late

This year's hiring advice for designers: retire the static case study, show clickable prototypes and video walkthroughs instead. Solid advice. It would have been ahead of the curve a few years ago.

The bar has already moved. The strongest thing a designer can hand over now is not a convincing picture of the product. It is the product, running. I found that out on a landing page for a business that would not tell me anything about itself.

A client who answered nothing

Over July and August I rebuilt the landing page for a port-to-door import logistics business. The site it had was a single WordPress page carrying roughly 31 words of product copy, substantively unchanged since January 2021. No address. No phone number. No email. No team, no customers, no statistics.

The engagement had a problem before it had a brief. Sixty-three questions went to the client across two rounds, covering the legal entity, whether the company was trading, what it actually sells, who it is for, and how to reach it. Sixty-three of sixty-three came back unanswered. Twelve of them were blocking.

So the design question was never "how do we present this business well". It was "what is a page allowed to say about a business that has told nobody anything, and how do we make that page persuasive anyway".

Pricing the silence

Waiting for sixty-three answers was not on offer. Guessing invisibly was not acceptable. The third path was to proceed on positions that are written down, owned, and cheap to reverse.

Every question went onto one severity ladder with a stable ID. Every load-bearing sentence on the page was tagged to the assumption it rests on, with that assumption's reversal cost: audience is a large reversal and stays flagged as blocking, spelling conventions are six strings and not worth waiting on. That separated the twelve questions that genuinely blocked anything from the fifty-one that did not, and let real work happen while the questions stayed open.

One direction, or three that share nothing

The normal shape of a website engagement is one design direction, refined through rounds of review. You present, they react, you revise, repeat until someone runs out of patience or budget.

I built three instead. Each complete. Each live on its own route. Each with its own tokens, components, stylesheet, behaviour script and data, sharing nothing at all. The token prefixes were deliberately different, so a rule copied from one direction into another fails loudly instead of half-working. Promoting any of them to production is a single line of configuration.

And they were genuinely different ideas, not three colourways of one layout. One is editorial and deliberately unfashionable: light paper, a serif display, one anchor hue and a signal colour used exactly once. One is a dark navy canvas built on a single variable grotesque, stretched through its width axis. One is a two-colour print job on bone stock, with tone made by halftone screens rather than a third ink, and crop marks on open elements.

The developer instinct is to share components across variants. I refused that too. The sharing is exactly what makes them variants of one idea rather than three ideas.

What it cost

Nothing propagates. A fix in one direction stays in that direction. Three complete directions mean three surfaces to keep correct instead of one, three sets of contrast checks, three sets of breakpoints.

That only makes sense because of the workflow. In an AI-assisted build, a second complete direction is nearly free, while unbounded variation on a single page is not. Free variation with nowhere to go turns into argument about one page. The isolation is what keeps the cheapness safe, and it turns the client's choice into a decision with a known cost instead of a negotiation.

One person, six roles, and a reviewer whose job is to say no

I covered every function myself, and the AI workflow is how one person covered that ground. The work ran through six named charters: project management, UX, UI, copy, frontend, and an adversarial QA reviewer. Each had its own required reading, its own scope, and an explicit instruction never to invent project facts.

The QA charter earned its place. It exists to send work back. Three rounds ran against the first direction, closing at zero blockers, zero majors and zero minors. Contrast was computed rather than eyeballed, 142 ratios published. And every automated gate was made to fail on a deliberately broken build before it was trusted, because the project had already shipped one gate that could not fail, and a gate that cannot fail is worse than no gate at all. It reads as coverage.

The honest part

The client later asked for a presentable page rather than a page with visible holes in it. The invented placeholder content the system had quarantined was exactly what would fill those holes, and the gate did its job: the first deployment attempt failed on sixty findings reaching the output.

Showing that content anyway was the client's decision. It was raised, reaffirmed, and recorded as a decision with the objection attached, not as an oversight. A gate can make invention visible, detectable and cheap to undo. It cannot make the call.

The deliverable was the thing

Here is the part the portfolio advice misses. The client never reviewed a prototype. They were handed three running pages, each on its own route, each with a design document saying what it is, what it took from its reference and what it deliberately did not. Choosing one is a single line of configuration, and the choice stays theirs.

The deliverable was not evidence that I can design. It was the thing.

If you are hiring a designer this year and they send you a beautiful clickable prototype, it is worth asking the follow-up question. Can they send you the running version instead?