How Do AI Robo-Advisor Alternatives Work?

A robo-advisor automates investing: it takes custody or discretion over an account and runs a model portfolio for you. An AI robo-advisor alternative does something different. It is usually not hands-off. It sits on top of a brokerage you already have, uses AI to help you understand what you own and what you might do, and leaves the decisions and the trades with you. The category spans a few shapes: chat-driven portfolio assistants like Walnut or Magnifi, second-opinion analyzers like PortfolioPilot, and tax-aware aggregators like Mezzi or Empower. They differ in emphasis, but they share the same backbone.

So the question “how does it work” is really four smaller questions: how does it connect to your accounts, how does it use AI, how does it help you decide, and how does anything actually get acted on. The rest of this guide takes them in that order, walking each step through Walnut as the worked example while pointing out where other tools do the same thing differently.

The first thing one of these tools does is link an account you already have, rather than opening a new one for you. It does that through an account aggregator: a service that securely connects to your brokerage and reads back your positions. Walnut, for example, connects your existing brokerage through a regulated aggregator; other tools in the category, such as PortfolioPilot, Mezzi, and Empower, link your accounts the same way, or through Plaid. You authorize the connection once, and the tool reads your positions from there.

The important detail is the permission level. In most well-designed alternatives the connection is read-only by default. The tool can see your holdings and prices, but it cannot move money on its own. With Walnut the link is read-only by default; trading, where supported, is a separate capability that you enable deliberately, and even then each order needs your approval. That read-only default is what makes connecting feel reasonable, and it is worth checking before you link anything, whichever tool you use. See whether it is safe to connect your brokerage to an AI for the full picture on access and security.

Once a tool can read your holdings, the AI part is mostly about context. The tool takes your real positions, usually adds live prices and recent web data, and hands all of that to a language model. Because the model is reasoning over what you actually own rather than a generic example, its answers are about your portfolio specifically. This is the step that separates the category from a plain chatbot, and every tool here does some version of it.

Concretely, that lets the tool produce plain-language analysis: how each holding has done, where you are concentrated, and how positions compare to a benchmark. Walnut frames each holding against the S&P 500 so you can see what is keeping pace and what is lagging; a second-opinion analyzer like PortfolioPilot instead rolls the same data into a risk score and a structured critique. Some tools let you pick the model you talk through. Walnut, for instance, lets you use Claude or ChatGPT or a built-in assistant, with web search for current information, where a tool like Magnifi leans on its own conversational layer.

One honest caveat: language models can still state wrong figures confidently, so a well-built tool grounds them in real data and you should verify anything specific before acting on it. Grounding the chat in your actual positions, the way Walnut does with your connected holdings, is exactly what separates this from pasting your holdings into a general chatbot. For that contrast, see the best AI finance chatbots comparison.

With your portfolio loaded and analyzed, the next step is conversation. You ask, in plain English, about what you own and what you are considering, and the model answers grounded in your real positions. This is where an AI alternative earns its keep: in Walnut you can pressure-test an idea, ask why a position is lagging the S&P 500, or explore a theme before committing to anything, and a tool like Magnifi plays a similar role when you are researching funds before you decide.

Crucially, the deciding stays with you. The tool can lay out trade-offs and frame how a change would look, but it is not running a model portfolio on your behalf. Some of these tools organize ideas into a theme or portfolio: a stated rationale plus a set of holdings and target weights. Walnut builds thematic portfolios exactly this way, so a research conversation can turn into something concrete and trackable rather than a one-off tip, while an analyzer like PortfolioPilot keeps its output closer to a report than a portfolio. Either way, you are the one drawing the conclusion.

The last step is where the alternative most clearly diverges from a hands-off robo. If you decide to act, the tool can build or suggest specific trades, but nothing executes until you approve it. The order is then placed at your own broker, so you keep the brokerage relationship and the final say. This is the part that defines the whole category: an alternative helps you act, it does not act for you.

With Walnut, connections are read-only by default, every trade requires your explicit approval, and the order is placed at the broker you already use. The tool helps you assemble and review the trade; you are the one who confirms it. Analysis-first tools like PortfolioPilot or Mezzi go a step lighter, stopping at suggestions and leaving you to place any trade at your broker separately. The shared principle is the same approval gate: an AI alternative is built to help you act, not to act for you.

Put together, the four steps form a single loop where you stay in control the whole way through. The table below summarizes who is doing what at each stage, with Walnut noted as one concrete illustration of each step.

The mechanics above add up to a different relationship with your money than a classic robo gives you. A few contrasts make the distinction concrete:

• Custody and discretion. A robo-advisor typically takes custody or discretion and trades on its own schedule. An AI alternative sits on top of your existing broker and does not take discretion; you place or approve the trades. Walnut, for instance, leaves your money at your broker and needs your approval for each order.
• Model portfolio vs your portfolio. A robo fits you into a preset model portfolio. An alternative works with the holdings you already have and helps you reason about them as they are, the way Walnut frames each of your real positions against the S&P 500.
• Hands-off vs hands-on. A robo is built to be set-and-forget. An alternative is deliberately not hands-off: it expects you to ask, decide, and approve. Walnut is not hands-off by design, and that is the trade.
• Performance framing. Because broker feeds rarely pass cost basis, a connected tool often frames performance as window returns (movement over a chosen period against a benchmark) rather than realized profit and loss, and says so. Walnut uses window returns for that reason.
• Advice posture. Most alternatives stay informational and descriptive. Walnut is not an investment adviser; it helps you research and frames holdings against the S&P 500, but the decisions are yours.

Neither model is automatically better; they suit different people, and Walnut is not the right fit for everyone. If you want true autopilot, a robo is the point. If you want to stay in the driver’s seat with better information, an AI alternative like Walnut, PortfolioPilot, or Magnifi is the shape of tool you are looking for. For the wider field, see the AI robo-advisor alternatives roundup, or the definition in what is an AI robo-advisor alternative.

To see the four steps end to end in one tool, here is how they chain together in Walnut (ours, so take the framing as a worked example rather than a verdict, and not a claim that it is best for everyone). You connect your existing brokerage, read-only by default. Walnut then analyzes what you hold, framing each position against the S&P 500. You discuss it in plain English through Claude, ChatGPT, or a built-in assistant with web search, turning a question into a thematic portfolio if an idea takes shape. If you choose to act, you approve every trade and it is placed at your own broker. That is the same connect, analyze, discuss, approve loop the rest of the guide describes, just in one place.

What Walnut is not: it is not hands-off, not a deep data terminal, and not an investment adviser. It frames returns as window returns because broker feeds rarely pass cost basis, and it has a free tier. It is one example of the flow, alongside PortfolioPilot, Magnifi, Mezzi, and others, not the only one, and the right pick depends on which part of a robo you are trying to escape.

Every mechanism above has a rough counterpart inside a managed account. One does not: the approval step. A robo-advisor decides and executes in the same motion, because that is what discretion means. An alternative assembles the orders and stops, and nothing happens until you look at them.

That pause is the whole product for some people and the reason others should not choose one. It is a genuine safeguard against an outcome you did not intend, and it is a genuine point of failure if you are the sort of person who leaves things in a queue.

AI robo-advisor alternatives work by connecting to a brokerage you already have (read-only by default through an aggregator), using a language model with your real holdings and web data as context to analyze your portfolio, letting you discuss it in plain language, and helping you build or approve trades that you place yourself. The throughline is control: unlike a hands-off robo that runs a model portfolio for you, an AI alternative keeps the decisions and the trades with you at every step. Walnut runs this exact loop, connecting the broker you already use, framing holdings against the S&P 500, and approving every trade with you, and it is one concrete example among several rather than the right answer for everyone. Walnut is not an investment adviser.