Joe da Silva · ProductAnywhere Concept gallery

Speculative design · four brief generations · 2026

Thinking,made inspectable.

Every design direction generated for this site, across frontier models, kept in the open — including the ones that missed. Each is a real, standalone HTML artefact with no build step. Open one and judge the thinking.

34artefacts
15models
4brief generations
3/3v5 models that refused to build a homepage

01

v5 — outcome brief · the models chose the artefact

The design specification was removed. These three were given the goal (get Joe matched to and hired into the highest-scoring roles), the evidence, the conversion data, and six hard constraints — then told that if a homepage was not the right instrument, not to build one. All three independently rejected the homepage and converged on a forwardable referral brief. Referrals convert ~11× inbound and carry 30–50% of hires; recruiters scan for 6–11 seconds, then read for 2–3 minutes.

02

v4 — ProductAnywhere · built on the evidence base

All generated from the same brief against the recovered evidence: 63 FormFunction client logos, 15 Remote Product speaker portraits, receipts carrying their stated limits. Every one passes the build gate — zero hex literals outside :root. Ten models, ten structural answers.

The hiring decision board

pick

GPT-5.3-Codex High · 18 Sep

Maps a hiring manager's actual risks — “can he lead beyond feature delivery?”, “can he work credibly with AI claims?” — to lane-coloured evidence. The only concept that argues against objections rather than around them. Strongest of the new batch.

open →

The dossier

pick

GLM-5.2 · z.ai · 18 Sep

Numbered index, exhibit-style receipts with confidence on the saturation ladder, burn-vs-typical chart built from the evidence. Most complete.

open →

The timeline spine

current

Kimi K3 · Moonshot · 18 Sep

“The years are the navigation.” Four chapters as a left rail with year watermarks. Sharpest hero copy.

open →

Asset-led editorial

current

Claude Opus 5 · 18 Sep

Logo wall as centrepiece, four-iteration panels, receipts with visible limits, speaker wall.

open →

Get Joe hired

current

Mistral Large · 18 Sep

Blunt, on-objective headline and a four-scale ledger. Plainest execution, clearest intent. All 63 marks inline.

open →

Operational architecture

current

Gemini 3.1 Pro · 18 Sep

Restrained and spacious: telemetry, operational history and architectural limits framed for a hiring mandate.

open →

Three-lane capability cards

current

Devstral 2512 · 18 Sep

Executive / Productized / Venture as equal cards with receipts beneath. Clean but reverts to the three-pillar layout the brief warned against.

open →

Spine & evidence

current

Qwen3-Coder · 18 Sep

Compact: the spine, then evidence-with-limits. Thin at 14 KB — least developed of the set.

open →

Minimal ledger

weak

Qwen3-Max · 18 Sep

Near-unstyled long-form ledger. Useful as a content skeleton, not as a design direction.

open →

— declined —

declined

Grok 4.20 · 18 Sep

Grok refused the brief, calling it too prescriptive, and offered a simpler scope instead. Kept as a record of the refusal.

open →

Phase-3 loop build

superseded

Kimi K3 · 20 Aug

The original loop's shipped home page. Predates the evidence recovery — two images total.

open →

Design system — tokens & ladders

partial

GLM-5.3 · 20 Aug

Stalled mid-phase-1: token file and two design-system sections. The token reference.

open →

03

Brand

04

v3 — regulated AI & portfolio redesign

05

v2 — three-pillar / “Portfolio of One”

06

v1 — original single-track brief

One direction per model, two passes each.

07

Briefs & specs