Vision

Democratize financial intelligence for all.

Markets are open. Judgment is not.

Everyone can see a price. Knowing what it means is harder. What to watch. What to ignore. When to act. How much to risk. And when to admit you're wrong. That judgment is earned across years and market cycles.

A top hedge fund wants a million-dollar minimum, often far more. A private bank wants a seven-figure relationship before it picks up the phone. The best investors spend their careers serving a small circle of wealthy clients and institutions, because that is where the fees are. Retail was never the customer. Not because the judgment stops working on a smaller account, but because serving it by hand doesn't pay.

Two sides of one problem

I have sat across from both sides of this. A trader I know has made more than $100 million trading US stocks. When I asked why he wasn't running more money, he said: “My time and energy are limited. Thirty years of results prove I can make money trading. But…” He didn't need to finish. He has the judgment and no hours. The retail investor has an account and no judgment.

Judgment is an idea, and ideas don't get used up. Paul Romer, who won the Nobel for working this out, put it plainly: “Once the cost of creating a new set of instructions has been incurred, the instructions can be used over and over again at no additional cost.” That trader's thirty years of pattern recognition would work as well on a hundred accounts as on one. What kept it scarce was never the judgment. It was the person it came attached to, with one body and twenty-four hours. Distilling it into an agent unbundles the two: the agent runs his know-how without getting tired, on more capital than he could watch alone, and brings that know-how into a retail investor's own account.

Reference: Paul Romer · Endogenous Technological Change · 1990

Public frameworks are a starting point. The harder asset is a professional investor's own judgment, and you don't get that from a dataset. You get it by working with them directly, one decision at a time, and by earning the right to keep asking. Models get cheaper every year. That kind of trust doesn't.

A great investor does more than pick a trade. They form a thesis, choose signals, size a position, and know what would prove them wrong. Most platforms show the final trade. We want to capture the process behind it.

We call this distilling the market mind: turning that judgment into something you can write down, test, and run.

Bridgewater AIA Labs, working with Thinking Machines, fine-tuned a model on expert investor labels. It outperformed tested frontier models on financial information filtering. Harvey's Tenet research preview uses expert data and legal task environments for reinforcement-learning post-training. That tells me specialized judgment can be learned. Whether it makes money is a separate question, and only the market answers it.

References: Thinking Machines × Bridgewater · Jun 2026 · Harvey Tenet · Aug 2026

1. Step 01Understand. Read the current state of the market.
2. Step 02Judge. Form a view of what comes next.
3. Step 03Act. Execute within defined risk limits.

An agent keeps that process running. It watches the thesis, reports what changed, and acts only inside the permissions it has been given.

Judgment needs a system that connects understanding to execution. Our financial harness brings together models, skills, plugins, and brokerages.

The four pillars converge on a customized, post-trained model for each professional investor. Retail investors use the same harness: their own agent copies a professional investor's trades within the limits they set, or simply trades for them.

The harness gives each agent portfolio context, risk limits, and a record of its decisions. We are the brain, not the vault. Users keep their assets in their own wallets and brokerage accounts.

An agent that leads has a job. It has a thesis to defend, risk limits it cannot cross, and copiers who will see every trade. If it talks well but loses money, the record shows it.

Every agent should earn trust in two stages: reproduce the judgment it was distilled from on unseen examples, then demonstrate results against a stated benchmark in live markets. You should be able to see returns after costs, the drawdowns, and the conditions under which the strategy stops working.

Return on Intelligence

Decision economics are part of that test. We call the expected value-to-cost multiple Return on Intelligence.

1× is break-even. Costs include incremental inference, data, and execution. As a hypothetical example, 5 basis points of pre-cost incremental edge on $4,000 is worth $2, just enough to cover a $2 total incremental cost. That is the number that makes a $4,000 account worth serving. No human advisor would take it.

Edge is uncertain and can decay as more capital follows it. The judgment doesn't wear out; the opportunity it trades can. Agents need to track both their cost and their capacity.

AI Finance Agent = Financial harness + Social. The harness makes judgment executable. The social layer makes it worth sharing.

Professional investors lead. Their know-how is distilled into an AI agent that trades on their judgment. Retail investors copy those trades with their own agent, inside their own accounts and risk limits. The link between the two sides is a profit share written into a smart contract: when copiers profit above a high-water mark, the professional investor earns a share. Settlement starts on-chain. That is how a professional investor puts more capital behind their judgment without managing anyone's money. And because the agent tracks its own capacity, copying can be capped before the edge decays.

The money moves in three lines, kept separate. Copiers pay the professional investor a profit share. Both sides pay Questflow for AI subscriptions and usage. Brokerage partnerships would add a share of execution fees. Nobody keeps paying for an agent that loses them money after costs, so the whole business rests on the same test as Return on Intelligence.

Crypto is our starting point, and not by accident. It trades around the clock, settles on-chain, and lets a smart contract pay a profit share without asking anyone's permission. The plan is one financial harness across US equities, Hong Kong equities, China A-shares, and beyond.

Our long-term ambition is an AI-native financial institution funded by durable operating earnings. After costs, partner shares, and taxes, profits retained in the company can support investment over many years.

Bill Ackman's Pershing Square Holdings is our reference for patient capital. Public shareholders exit by selling shares in the market; they cannot demand redemption from the company. This reduces pressure to sell portfolio assets to meet investor withdrawals, giving capital time to stay invested through market cycles.

Berkshire Hathaway illustrates the operating model. Cash generated by See's Candies helped fund other businesses, whose profits could then be allocated again. The holding company can invest across operating businesses and marketable securities.

For Questflow, the plan is a holding company that takes the cash the operating businesses throw off and puts it back to work: into the harness, into businesses that generate more cash, and into long-term investments. The structure buys time. What we earn on it comes down to how well we allocate. Client assets stay separate from all of this.

Charlie Munger put the whole idea in one line: “The first rule of compounding is to never interrupt it unnecessarily.” Permanent capital is how we plan to obey it.

References: Bill Ackman · PSH 2023 Annual Report · PSH share structure · Berkshire · 2014 shareholder letter · Munger · Poor Charlie’s Almanack

None of this works fast. Judgment has to be distilled one investor at a time, proven on unseen decisions, then proven again with real money against a benchmark. Copiers have to make money after costs before they keep paying. Only then is there cash to retain, and only retained cash compounds.

That suits me. Everything else in AI is a race. Models get replaced every few months, and whatever we build on top of them will get rebuilt. The one thing in this business that is allowed to go slow, and actually gets better by going slow, is the compounding of money. I want to protect that part.

A finance agent is also the only kind of agent I can think of that you can run as an infinite game. It never ships and finishes. It keeps a P&L every day the markets are open, and a very small team can run it for decades with almost infinite leverage: the agents do the work, and the capital compounds. That is how we intend to play: long-term games with long-term people, as Naval put it.

So we are building for decades, and the holding company above is what lets us.

That is the work of my life. Financial intelligence for all.

Founder, Questflow