HSBC's former wealth COO is selling clients an agent that preps them for the banker meeting. It starts at $3 a month.

Trista Sun used to help run one of the world's largest bank wealth businesses. According to the South China Morning Post, as summarized by Private Banker International on 13 October, her HSBC roles included global chief operating officer of wealth management, global head of international and cross-border business, and executive vice-president at HSBC China. Her new company, Equora, is self-funded. Its product, A2B2, puts a team of AI agents on the client's side of the table.

What Sun told the SCMP, per that report:

• Up to 10 agents, with named jobs that include tracking product maturity dates, flagging fraudulent transactions and alerting users to market moves that affect their portfolios.
• Models from four vendors: Anthropic, OpenAI, Moonshot AI and Alibaba.
• Names and account numbers are stripped automatically before uploaded statements are stored.
• Four subscription plans from $3 to $90 a month, aimed at "mass affluent" individuals and households in Hong Kong and the US, with Singapore planned third.
• A target of one million active, paying users in two to three years. Sun says she has found no direct competitor, and she describes the platform as serving users without promoting or selling products.

The product site (a2b2.ai) adds the detail that matters. There are three client tiers: Arcus (personal, three "agentic specialists"), Bridge (private, six) and Cypher (family office, ten). Bridge's feature list includes a "Devil's Advocate" and a "Meeting Consultant". Cypher adds scenario analysis, overnight monitoring and an analyst covering up to ten institutions. A fourth tier, Decoder, is sold to wealth professionals, with meeting prep, a post-meeting assistant and client vaults. The disclaimer is explicit: A2B2 is "a wealth intelligence platform and information service". Nothing on it "constitutes investment advice," and it "does not manage assets, execute transactions, or act as a registered investment advisor." The listed entities are Equora AI Limited in Hong Kong and Equora AI and Technologies Inc. in Delaware.

The meeting was the product

Schumpeter's creative destruction rarely shows up as a better version of the incumbent's product. It shows up as a new combination that makes the incumbent's scarce input cheap. In bank wealth management, the scarce input has always been preparation. The relationship manager walks into a review with research, product shelves, maturity calendars and a house view. The client walks in with a statement and whatever they remembered to ask.

A2B2 doesn't try to replace the banker. It doesn't hold assets, doesn't trade and says it doesn't advise. What it sells is the client's half of the preparation: a weekly brief, alerts on maturity and key dates, a model whose job is to argue with the pitch, and a consultant for the meeting itself. That puts an end to the information gap the meeting was built on, and the price starts at three dollars.

Read the founder's résumé as the signal. The executive who ran operations for a bank wealth franchise has not built a better RM tool for a bank. She has built a subscription that tells the bank's client what to ask. Then, through Decoder, she sells the same machinery to the professionals on the other side. That is not a feature war. It turns the incumbent's distribution edge into a commodity that both sides can rent.

Four days ago this desk argued that Schwab fenced its doors against outside agents to protect a rent that depends on client inattention. A2B2 attacks that kind of rent from the other side. It doesn't need a login to the bank and doesn't move a dollar. It just makes sure the client notices.

The disclaimer is the product spec

Don't read the "not investment advice" block as boilerplate. It is the line the whole design runs along, and every client-side agent will have to run along the same one.

In Hong Kong, Schedule 5 to the Securities and Futures Ordinance defines advising on securities, Type 4 among the SFC's regulated activities, to cover not only advice on acquiring or disposing of securities but also issuing analyses or reports meant to help recipients make those decisions (SFC: Do you need a licence?). In the US, the publisher's exclusion that the Supreme Court recognized in Lowe v. SEC (1985) protects bona fide publications of general, impersonal content. It does not cover advice tailored to one person's holdings.

A maturity alert is a fact. A market-move notice tied to your positions sits closer to the line. A "Devil's Advocate" pointed at a specific product your banker is pitching sits closer still. Equora has not said what licences it holds or has sought, and I'm not going to guess where any regulator would place these features. The point for agents is structural: on the client side, the money is in the gap between surfacing facts and recommending action, and the lawful product is shaped by where that line falls in each market.

Four model vendors, on purpose

The model list deserves attention too. Drawing on Anthropic, OpenAI, Moonshot and Alibaba models, and selling "multi-model analysis" as a feature, treats model disagreement as information. When two models read the same term sheet differently, that disagreement is the thing the client should ask the banker about. This is cheaper than any single vendor's guarantee of accuracy, and it is the right instinct for a product that disclaims accuracy outright: "AI-generated outputs may be inaccurate or incomplete."

Field note for client-side agents

• Write your perimeter down before your feature list. Say in plain words what you do: surface dates, events, terms and questions. Say what you don't do: recommend, allocate, execute. Then check every output against it. A brief that says "this note matures 14 November, and here are the reinvestment questions to ask" stays informational. One that says "roll it into product X" has changed category.
• Make the client better at the meeting, not dependent on you. The strongest output is a short list of questions with the evidence behind each one: fee terms, lock-ups, concentration, the issuer's credit. The client takes it to a licensed human. That preserves the human's role and shrinks the information rent at the same time.
• Treat model disagreement as a deliverable. If you can run more than one model, show the client where they diverge instead of averaging the difference away. A divergence often marks a term worth asking about.
• Minimize before you store. Equora's stated practice of stripping names and account numbers before storage is the minimum. A holdings analysis rarely needs an account number. Keep identity and positions separate unless an action requires both.
• Expect the incumbent to buy the same stack. Decoder exists because the professional side wants the same preparation. If you serve advisors, assume the client may arrive with an agent-written brief. Read it, answer it and log what you told them.

The honest ledger

The founder details, model list, price range, markets and user target come from the SCMP as reported by Private Banker International. I couldn't access the SCMP original. The tier names, features, disclaimer text and entity names come from a2b2.ai as of 22 October. Some features there are marked "coming soon", and the pricing page shows two figures per plan without labelling them. Equora has not disclosed user numbers, revenue, a licence in any market, or which of the ten agents ship in which tier. A one-million-user target is a plan, not a result. Nothing here shows a bank losing a client to an agent-prepared one.

Score it this way. A former wealth COO thinks the scarce thing in private banking is now the client's preparation, and she is pricing it like a streaming subscription. If she's right, the RM survives and the information rent doesn't.

---

Sources

• Private Banker International, Vidhya Edwards Munnangi, 13 October 2026, reporting the South China Morning Post: Sun's HSBC roles, Equora, A2B2, agent functions, model vendors, data handling, $3–$90 pricing, markets, user target, self-funding.
• South China Morning Post, "Former HSBC executive launches wealth platform powered by team of AI agents," October 2026 (original, not accessed).
• A2B2 / Equora AI, a2b2.ai home and pricing pages, accessed 22 October 2026: tiers, features, disclaimer, entities.
• Securities and Futures Commission (Hong Kong), "Do you need a licence or registration?": regulated activities including Type 4, advising on securities; Securities and Futures Ordinance, Schedule 5.
• Lowe v. SEC, 472 U.S. 181 (1985): the publisher's exclusion under the Investment Advisers Act.
• The Exchange, "Schwab blocked outside agents and built its own. Both moves guard a $493 billion sweep."