As buying questions move to AI assistants, this store is already built to be their answer. AI agents rebuilt a ten-year e-commerce operation for the 2026 season, structured for AI search from the catalogue up.

The 2026 product line started with a documented design language: a cast of more than 25 original characters, a colour law governing palettes, a pattern grammar drawn from traditional knitwear design, and a reusable library of named motifs. AI image generation produced the artwork; every design passed a scripted quality inspection covering colour, pattern, and layout rules before the owner saw it, and nothing published without her approval.

Twenty-one designs made the season. From those, agents built the full product matrix: all-over-print garments, 49 tee products across adult, kids, and city-fan lines, 40 traditional knitwear products, 65 matching sets pairing people with their partners, kids, and dogs, a pet line, and a custom studio where customers submit their own design for a made-to-order garment. Sets carry up to 63 size combinations each, and pricing is size-inclusive by policy: one price per product, every size. A companion line of 13 party goods runs on a print marketplace under the same brand.

Cart logic handles the merchandising a category manager would: matching sets cross-sell their missing piece, and a family bundle price applies automatically when a household's pieces share the same pattern.

The content operation covers the season rather than the week. Thirty-one legacy articles were rewritten to current standard: expanded to 1,200 to 1,600 words, given FAQ and verdict sections, re-linked, and re-monetised. Seventy-three new articles were written, checked against a house style law, and scheduled from July through mid-December, timed so that December's search content is live by the end of September. Publishing runs on WordPress's own scheduler, so the pipeline needs no AI involvement on the day.

Built for AI search as well as Google. A growing share of buying questions is asked of AI assistants rather than a search bar, and an answer engine can only recommend what it can read. The site was built for that reader from the catalogue up. An llms.txt guide file tells AI crawlers what the site is and where its substance lives. Entity schema establishes who the brand is. Structured data runs through the whole estate: 208 FAQ answers across 34 articles, written in the question-and-answer form assistants retrieve, product schema on 47 products so price, availability, and sizing are machine-readable, and list schema on 16 gift guides so roundup recommendations resolve to specific products. Traditional search gets the same layer through XML sitemaps and on-page optimisation, so one build serves both kinds of reader.

The server logs show the AI readers arriving. In a single three-day window in July, months before the season peak, AI crawlers made more than 900 requests to the site: Anthropic's ClaudeBot, OpenAI's search crawler and GPTBot, and the AI crawlers run by Meta, Amazon, and TikTok. Thirteen of those requests came from ChatGPT fetching pages in real time to answer a user's question. The store is being read by the assistants now, positioned to be cited and recommended as the season's buying questions arrive.

Social and video. More than 150 social posts sit scheduled in the calendar, one per article plus campaign series, across Facebook, Instagram, Pinterest, and Threads, running from July to late December. Six platforms publish from a single social planner inside the CRM, so scheduling is one integration rather than six. The video program produced 22 short-form videos with AI video generation: a launch teaser plus a deadpan character series covering the full 21-design cast. Each short is scheduled twice, to TikTok and to YouTube Shorts on mirrored calendars. A daily automated watchdog sweeps the posting calendar for failed posts, expired platform tokens, and stalled schedules, and repairs the known failure modes without waiting for a human to notice.

Email and CRM. Contacts live in a segmented CRM with tag-based lifecycle tracking. Every new contact enters a four-stage welcome sequence, then graduates to a four-stage nurture sequence at days 14, 21, 30, and 45, covering matching sets, the custom studio, the character cast, and shipping offers. Abandoned checkouts get their own recovery sequence. Suppression, unsubscribe handling, and a one-email-per-day cap are enforced in code, and the responsive branded templates render correctly down to narrow inbox panes. The sending logic runs on the site itself, so the sequences operate with no third-party automation to maintain.

Channels and integrations. The store connects outward through live product feeds: Google Merchant Center syncing the catalogue for Shopping listings, a Pinterest business integration with conversion tag, server-side tracking, and an 8,297-variant catalogue feed, and a Microsoft Advertising account staged to import the Google feed. Checkout offers five payment methods: Stripe, PayPal, Klarna, Affirm, and Link. A tiered affiliate program runs on the store, analytics run on GA4, and a corporate channel handles bulk and event-planner enquiries. Every integration was built and verified by agents through APIs: the WordPress REST API for content and catalogue, the Printful API for products and fulfilment, the CRM's API for contacts, email, and social scheduling, and the platform connections above.

How it runs between humans. Six scheduled agent routines operate the business: a daily commerce health check across the site, payment gateways, and order forwarding; a Monday content session that keeps the publishing pipeline at least four posts deep; a Friday report email to the owner with the week's numbers and anything needing her attention; a monthly full audit of links, plugins, and the catalogue against the seasonal calendar; the daily social watchdog; and an hourly sweep that archives every generated design asset. Alerts go out only when something needs fixing. The owner's standing routine is reading the Friday email. Decisions that need her judgment, and anything outward-facing or irreversible, queue for her explicit approval. She checks in on the business roughly every six weeks.

Ten years of continuous trading, over 80,000 unique visitors in the most recent peak season, and a 2026 season rebuilt end to end by AI agents: 180+ products, 100+ articles, 150+ scheduled social posts, 22 videos, an eight-stage email lifecycle, three shopping-channel integrations, and a structured-data layer built for AI-assistant search, delivered without a development, design, content, or production team. A decade of operation demonstrates that a business built on AI tools can sustain real commercial trading at real scale; the 2026 rebuild demonstrates what the current generation of AI agents can execute against a live business in weeks.

Every tool Southern Sky AI recommends to a client is already operating in a live commercial context. The AI agents, the automation logic, the design governance, and the build-without-a-team approach described here are the same ones Southern Sky AI recommends to clients, running with real orders, real customers, and real revenue on the line. What gets recommended is what already works.