Shirley XueAll work
ablefyVision DemoAI-native2026

A working console for an AI-operated platform

Problem
The strategic question — what do we look like when AI operates the platform — was being answered with decks.
Solution
A working console: the money flow visible, identity verification (KYC) as a conversation, and a co-pilot that proposes moves the seller approves.
Result
Vision demo for internal alignment, on synthetic data — not a production rollout.

Role & team

Strategy, design and build — solo

  • Owned the strategy, designed the console, and coded the front end by directing AI coding agents.
  • Built across three repositories.

Status

Runs end to end

  • Console, community satellite and agent-payment layer.
  • The seller, her products and every number on screen are synthetic.
  • Stakeholders click through the whole story instead of reading a deck.

Safeguards

The AI suggests; the seller decides

  • Every suggestion is labeled as one, and nothing acts without a click.
  • Automated assertions hold the co-pilot's logic.
Context

At ablefy, a creator-commerce platform, I answered that question with software: a console demo where the platform's backbone (checkout, invoicing and VAT, payouts, verification) runs underneath the seller's connected products, and an AI co-pilot on top watches everything and proposes the next move.

Backbone

One backbone, a constellation of products

The platform is not one app but the backbone under many. A seller's community, course studio and even a third-party ticketing product connect to it as satellites, each borrowing checkout, invoicing and VAT, payouts and KYC. The console is the seller's one place: pending payout top left, satellites with their revenue in the center, and the co-pilot's current suggestion visible without scrolling.

The console home — pending payout, an AI suggestion card labeled 'AI suggestion, your call' proposing an annual community plan, and a constellation of three connected products with their 30-day revenue and ordersView full size
Home (synthetic seller and numbers) · the backbone, the satellites, and a co-pilot suggestion labeled as one
Money flow

The money flow, end to end

The most future-facing order in the demo costs €0.10: an AI agent buying a per-question consult on a buyer's behalf. It sits in the same feed as a €299 course sale, because to the backbone an agent purchase is just another order, with the same invoice, VAT and payout obligations. Every sale from every satellite lands in that one feed.

Money logic stays deterministic and visible. The payout view shows what is accruing, when it pays out, and net, VAT and gross for every processed payout. The platform is merchant of record (the legal seller toward the buyer), so the seller never touches VAT, buyer invoices or payment reminders.

The orders feed — sales from three different satellite products in one list, including an AI agent's 0.10 euro per-question purchase, each with product chip, price and PAID statusView full size
Orders · every satellite's sales in one feed, including the AI agent's €0.10 purchase
The payouts view — accruing amount for the next payout, the scheduled payout date, and processed payouts each showing net, VAT and grossView full size
Payouts · net of VAT and fees, scheduled automatically, reconciled on screen
Smart KYC

Verification is a conversation, then a watched state

Compliance onboarding is where platforms lose sellers. Getting verified is a conversation with an agent instead of a pile of forms. Staying verified is the backbone's job: identity document, payout account and VAT registration are watched and re-checked automatically.

Smart KYC — a Verified banner for the demo seller, and three watched items (identity document, payout account, VAT registration) each marked verified with plain-language explanationsView full size
Smart KYC · the verified state, with three watched items
Co-pilot

The co-pilot proposes moves the seller approves

It watches everything the backbone sees (orders, invoices, recovery, every satellite) and turns it into one-click moves: an annual plan that would cut churn, a course that would package existing knowledge, a payment plan that would lift a product's lifetime value, a nudge for students stalling at module three. Every card carries its reason and its money consequence, and every action routes into the same tools the seller could have used by hand.

The design rule carried over from my other AI work: the model proposes, and the numbers on the card come from the deterministic backbone.

The business co-pilot — a week-at-a-glance KPI strip and a 'what I'm seeing' list of four suggestion cards, each with a plain-language reason and a one-click action like 'Price it with the agent' or 'Nudge via your AI coach'View full size
The co-pilot · suggestions with reasons and money consequences, one click from acting, none acting on its own
Outcome

What came of it

The backbone framing gave the platform's agentic direction a concrete shape that people could critique.