AI Investing Without Automatic Trading: Tools That Recommend but Do Not Trade

Discretionary means someone has authority to trade your account without checking with you first. Non-discretionary means every trade needs your say-so. That is the whole distinction, and it is a legal one about authority rather than a technical one about how capable the software is. A very sophisticated tool can be non-discretionary and a very simple one can be discretionary.

It matters because it is the line between a tool that can be wrong and a tool that can be wrong and act on it. If an AI misreads your position and tells you something incorrect, you notice and ignore it. If it misreads your position and rebalances, you find out later, and undoing it may be a taxable event. The quality of the model does not change that asymmetry. Only the authority does.

There is a second line worth knowing, which is where the account lives. A robo-advisor opens an account, holds your money and manages it. A connected app links the account you already have and reads it. Those are usually bundled together, so products that hold your money also tend to have discretion and products that connect to your broker tend not to, but they are separate properties and it is worth checking both.

One thing to be clear about, because it comes up constantly: a read-only brokerage connection cannot place an order. Not “is designed not to”, but cannot, because the access it holds does not carry the permission to trade. Whether a connection is read-only or trade-enabled is set when you authorise it, and it is worth knowing which one you agreed to.

This is a regulated distinction rather than a marketing one, so the primary sources are worth reading directly.

• SEC, Staff Bulletin: Standards of Conduct for Broker-Dealers and Investment Advisers, Care Obligations. Read it. Sets out the care obligations that apply to broker-dealers under Regulation Best Interest and to investment advisers under the fiduciary standard of the Investment Advisers Act of 1940. Both are drawn from fiduciary principles that include acting in the retail investor's best interest and not placing the firm's interests ahead of the investor's.
• Investor.gov (SEC Office of Investor Education and Advocacy), Robo-Adviser. Read it. Defines a robo-adviser as an automated digital investment advisory program that collects your goals, horizon, income, assets and risk tolerance through an online questionnaire, then creates and manages a portfolio for you, often at lower cost than a traditional advisory programme.
• FINRA, with the SEC and NASAA, Artificial Intelligence (AI) and Investment Fraud. Read it. A joint investor alert, published January 2024, warning that bad actors exploit the popularity of AI to lure victims, and that claims about proprietary AI producing guaranteed or outsized returns are a recognised fraud pattern.

Two things follow for anyone choosing a tool. Whether a provider is a registered investment adviser decides whether anybody owes you a duty of care, and it is a question with a checkable answer. And any claim that a proprietary AI produces guaranteed or reliably market-beating returns is the pattern the regulators have specifically warned about, whatever else is true about the product.

The safest version of the answer, in the narrow sense that there is no mechanism for an unwanted order because there is no connection to a brokerage account.

The shared limitation is the important one. Neither can see what you actually own, so neither can answer the question that most often matters, which is not “is this a good stock” but “is this a good stock for me, given what I already hold”. A stock can be excellent and still be a bad addition to a portfolio that is already 40% weighted to its sector.

This is the group the question is usually reaching for: an AI that can see your real positions, and therefore say something specific about them, without holding the authority to act. The account stays where it is, the connection is read-only unless you change it, and any order is assembled for you to approve.

The three differ more than the shared category suggests. PortfolioPilot is closest to a periodic critique: it looks at everything and returns an opinion. Magnifi is closest to a research surface, strongest when you are discovering funds rather than reviewing what you hold. Walnut is closest to an ongoing conversation about the specific positions in an account you already have. If you are choosing between them, the useful question is whether you want a verdict, a search tool, or a conversation.

These are not failures of the category, they are a different product, and they are here because a page about tools that do not trade is not much use without showing what the other side looks like.

Worth saying plainly: discretion is not a trap, it is a service, and for a lot of people it is the right one. Someone who knows they will not rebalance, and knows they might panic-sell in a drawdown, is often better served by a robo-advisor that simply does it. The mistake is not choosing discretion. The mistake is choosing it without noticing.

This is checkable in a few minutes and almost nobody does it. Four questions, in order of how much they matter.

• Is the connection read-only or trade-enabled? Your broker usually shows this in its connected-apps or third-party-access settings, and the tool should state it too. If the two disagree, believe the broker.
• Can any order be placed without an explicit approval step? Ask it about the edge cases: scheduled contributions, rebalancing, anything described as automatic. The interesting answer is never about the normal path.
• Is the provider a registered investment adviser? This decides whether anyone owes you a duty of care, and it is a matter of public record rather than something to take on trust from a marketing page.
• How do you revoke access? Find the button before you need it. Revocation lives with your broker as well as with the tool, and the broker's version is the one that actually cuts the connection.

If you want to see how this looks in practice, our guide to whether it is safe to connect your brokerage to an AI walks the same ground from the security side, and which brokers have an AI assistant covers what individual brokers currently expose.

Walnut sits in the approval-required group deliberately. It connects the brokerage account you already have, read-only by default, so it can answer questions about your actual positions rather than a description of them. Where a broker supports trading through the connection, Walnut assembles the orders and shows you what they are; nothing is placed until you approve it. Your money never moves to Walnut, because there is nowhere for it to move to.

The honest limits. It will not manage money for you, so if what you want is to stop thinking about it, a robo-advisor is the better answer and we would rather say so. Most brokerage connections are read-only anyway, so for a lot of accounts the trading question is moot. It does not model retirement income, withdrawals or Social Security. It frames returns as window returns rather than lifetime profit and loss, because most brokerage connections do not expose cost basis. And it is an informational tool, not a registered investment adviser and not a fiduciary.

Start from how much authority you want to give away, not from which AI is best, because the second question only matters after the first is settled.

• You want to think, not delegate. A general assistant is enough if the questions are conceptual. A connected app is better the moment the question involves what you actually hold.
• You want a second opinion on real positions, and you will place the trades. The approval-required group, chosen on whether you want a verdict, a research surface, or a conversation.
• You want the job done without you. A robo-advisor, and grant the discretion knowingly. Automation is the product you are buying.
• You have a systematic strategy already. Strategy automation, where per-trade approval would defeat the purpose.

Nothing here is a recommendation about what you should own. It is a description of how these products differ on the one axis that is hard to reverse.