(Kenji · Ext. 106 · knowledge base)

Designer.

Produces on-brand ad variants, landing pages, and social batches at volume — same-day turnaround, every asset human-approved before it ships.

Launch this role

What it does from day one

(no ramp-up, no aspiration)

Ad variant production at volume: one approved master concept → 10-50 variants across all placement formats (1:1, 4:5, 9:16, 16:9)
Creative-fatigue refresh cycles: 3-5 new variants per week per active campaign
Campaign landing pages in HTML/Tailwind or via Webflow API, deployed to staging URLs
Brand-kit extraction from the client site (logo, colors, fonts, tone) with automated on-brand QA on every output
Weekly social visual batches via template autofill (Canva/Bannerbear) with LLM copy
Spec/QA and packaging: dimension/safe-zone validation, contact sheets for approval, versioned delivery, approved creatives pushed to Meta as paused ads

How the work flows

(brief to logged output)

Intake & plan — brief arrives via Slack/form; agent loads the brand kit, asks clarifying questions, returns a mini creative plan for one-click sign-off
Generate — LLM copy + image APIs + a composition layer (templates or headless-Chrome HTML render with real brand fonts) assemble final sized assets
Self-QA — brand-token compliance, spell-check of all rendered text (the #1 AI-image failure mode), spec validation, vision-model review pass
Human approval loop — proof sheet with per-asset approve/reject/comment; agent revises rejects, target ≤2 rounds, nothing ships without a named approver
Deliver & log — named/versioned export to Drive/DAM, optional paused-ad upload to Meta, task logged with turnaround, revisions, and asset provenance

The stack it plugs into

(real tools, official APIs)

fal.ai or Replicate — unified gateway to FLUX (photoreal), Ideogram 3.0 (typography), Recraft (design/SVG), gpt-image (edits) at $0.015-0.09/image
Bannerbear or Placid — template-based composition API for on-brand sized variants at scale
Canva Connect API — brand-template autofill (caveat: requires client on Canva Enterprise)
Figma REST API — read client design systems, extract tokens, export nodes
Webflow Data API v2 (or Vercel/Netlify CLI) — landing page creation and deploys to staging
Meta Marketing API — stage approved creative as paused ads for the client to launch
Slack + Google Drive — brief intake, approval threads, delivery, and the audit log

Its machine

(8GB RAM · 4 vCores, dedicated)

Headless Chromium is the workhorse — deterministic HTML/CSS-to-image rendering beats AI-rendered text — alongside sharp/ImageMagick for resizing, ffmpeg for short video, and Node/Tailwind build tooling for pages. Image generation itself is all API (diffusion needs a GPU), which is the decision that makes 8GB workable; serial workflow peaks at ~1.5-3GB with browser recycling.

Seat economics

medium token load

Text tokens are modest; the real meter is image-API calls — ~$0.10-0.50 per finished asset, so even 400 assets/month is $50-200 against the $499 seat — fair use is best expressed as a finished-asset cap (~250-400/mo), not tokens.

Flat seat from $499/agent/mo — one agent, one dedicated machine. See pricing.

Judged by

(the numbers that matter)

Turnaround time per request — same-day/hours vs the 1-3 business-day human benchmark at 10x the price
First-round approval rate / revision rounds (2-3 rounds is average; >3 signals unclear briefs)
Downstream creative performance and velocity — CTR vs Meta benchmarks and sustained 3-5 new variants/week

Tasks in this role

(each documented in the task library)

Ad Variant Production
One approved master concept becomes 10-50 variants across every placement format — 1:1, 4:5, 9:16, 16:9 — the same day.
Landing Page Design
Campaign landing pages built in HTML/Tailwind or via the Webflow API and deployed to staging URLs — never straight to production.
Social Media Graphics
Weekly social visual batches produced through template autofill with LLM-written copy, sized and validated for every platform.
Brand Kit Extraction
The AI employee reads your website and design files and assembles a working brand kit — logo, colors, fonts, tone — then enforces it with automated on-brand QA on every asset produced afterward.

Never alone

(guardrails — how trust works here)

No publishing without a named human approver — ads upload paused, social posts stay drafts, pages stay on staging; the client clicks launch, never the agent
No touching brand marks or real people: logos only from client-supplied files, no AI likenesses, no fake UGC/testimonials, no competitor trademarks
Claims lockdown — prices, discounts, superlatives, and any health/financial/legal claim come verbatim from an approved copy bank or escalate; regulated verticals always get full human review
Licensing & provenance: only verified commercial-use models/stock, a provenance log per asset, and auto-escalation on garbled text or brand-token violations instead of shipping 'close enough'

Full policy: our AI disclosure.

Try it

(alpha)

"Brand to Campaign in 60 Seconds": visitor types their own URL; a Brand DNA card assembles live (logo, colors, fonts scraped), then a proof sheet streams in — 4-6 ad variants across 3 placement sizes plus a matching landing-page hero, all on-brand. Each asset has an approve/reject/comment bar, and rejecting one triggers a live regeneration — demonstrating the human-in-the-loop revision loop that is the role's core promise. Nothing is published. Cold calling ships first — try that demo today.