What Is Agentic Trading, and What It Is Not

The phrase "agentic trading" appears in product pages, conference talks, and financial technology coverage with increasing frequency. It is used to describe systems that range from disciplined, rules-based portfolio engines to tools that claim to forecast market moves and trade on those forecasts. Those two things are not the same. Knowing which category a system belongs to is one of the most important checks a retail investor can run before giving any software access to a live account.

What the word "agent" means in software

In software engineering, an agent is a system that perceives its environment, takes an action based on what it observes, and repeats. The defining characteristic is the loop: read, decide, act, repeat. The goal the agent works toward is defined by whoever built it, not chosen by the agent itself.

In the context of a portfolio, the environment is your live positions and their current market values. The action is a set of trades. The goal is a target allocation defined in advance, using a published method. The agent runs the loop on a schedule, or when a trigger fires, and its output is the set of orders that would bring the live portfolio closest to the target.

This is a precise and testable thing. You can inspect the method, verify the inputs, and check whether the output is consistent with the stated rule. A reader who has worked through portfolio optimization versus rebalancing will recognise this loop: it is the same process applied automatically rather than by hand.

The two things "agentic trading" is used to mean

The term gets applied to two categorically different systems, and the marketing rarely distinguishes them.

The first is a rules-based portfolio agent. It holds a target allocation derived from a published method. It reads live positions. It calculates the drift between what is held and what the method says should be held. It proposes the trades that close that gap. The method does not change unless the investor changes their inputs. The agent does not form opinions about where individual assets are going.

The second is a system that claims to forecast market behaviour and trade on those forecasts. It predicts, or claims to predict, which assets will rise or fall over some horizon, and places orders based on those predictions. The logic is often described as an AI that finds opportunities or learns market patterns. The track record of systems in this category, taken as a whole, is difficult to verify externally, because the forecast logic is proprietary.

These two systems have different risk profiles, different transparency standards, and different relationships with the investor. Using the same category label for both obscures a distinction that matters before you act.

What a rules-based portfolio agent does in practice

A rules-based agent runs a defined cycle. The cycle starts by reading the current state: what positions are open, what they are worth at current prices, and what fraction of the managed capital each represents. It then re-derives the target allocation using the published construction method applied to current market data. The result is a set of weights, one per asset, that represents where the portfolio should be according to the method.

The agent then compares the two: the live weights and the target weights. Where the gap exceeds a threshold, it calculates the orders required to close it. It does not decide whether those orders will be profitable. It applies the method. The method handles risk distribution, correlation, and position sizing. The agent handles the comparison and the execution proposal.

This is the process described in detail at what an AI portfolio agent is: a continuous loop operating on live data according to a stated rule, not a system making discretionary bets. The agent is only as good as the method it applies, and the method is the part you can read before you connect anything.

What agentic trading is not

A rules-based portfolio agent does not predict short-term price movements. It does not know whether the market will rise or fall tomorrow. It does not have a view on individual stocks. The construction method it applies, whether hierarchical risk parity or mean-variance optimisation or a comparable approach, is a risk-distribution technique, not a forecast. It allocates capital in a way designed to be robust across scenarios rather than optimised for a single predicted outcome.

It does not exercise discretion. It cannot decide that a particular asset looks overvalued and reduce its weight beyond what the method prescribes. The investor defines the universe and the constraints; the method defines the weights; the agent applies both. Any deviation from the method requires changing the inputs, which requires the investor to act.

It does not claim to beat the market. A system that uses the language of risk-adjusted allocation and drift management is telling you something precise about what it does. A system that implies it can predict what will happen is making a claim that no rules-based agent can support, because rules-based agents do not make predictions. Acubic publishes no track record and makes no prediction about future performance. That is not a disclaimer added for legal reasons; it is an accurate description of what a construction method does and does not know.

Why this distinction matters before you act

The practical consequence is straightforward. A rules-based agent has behaviour you can anticipate. You know what method it will apply, what inputs it uses, and what conditions would cause it to propose a trade. You can read the published construction method and understand what the system will do before you connect a live account. The risk you are taking is the risk of the method, which is a known and stated quantity.

A forecasting system has behaviour you cannot anticipate in the same way, because the forecast is the unknown. The system may perform well for a period and poorly for another, and there is no structural reason to expect it will outperform over your specific holding period. The risk you are taking includes both the method risk and the risk that the forecast is wrong, and the second of those is not bounded by any published limit.

For a retail investor choosing where to apply automation, knowing which category a tool belongs to is a prerequisite for any other evaluation. The technology label and the interface can look identical. The underlying commitment is not.

What to look for when evaluating an agentic portfolio system

A rules-based system should be able to answer the following questions clearly. If it cannot, that is a signal worth weighing before connecting real money to it.

• What is the published construction method, and where can you read it in full?
• What inputs does the method use, and how are they sourced and updated?
• What happens when the method proposes a trade: is there an approval step, or does the system act immediately?
• What are the stated limitations of the method, and what scenarios is it not designed to handle?
• How can you disconnect, and what happens to your positions when you do?

A forecasting system typically cannot answer the first two questions with verifiable detail, because the predictive logic is proprietary. That opacity is worth treating as a material fact rather than a feature.

The article how Acubic works covers each of these questions in sequence. The construction process, the approval step, and the stated limitations are published before you connect an account, not after.

A category that needs a more precise vocabulary

The practical argument for a rules-based approach is not that forecasting is impossible. It is that rules-based behaviour is verifiable, its failure modes are bounded, and its output is auditable. You can check whether the agent did what the method says it should have done. You cannot check whether a forecast was reasonable after the fact; outcomes are not a reliable guide to the quality of a prediction made before them.

Agentic trading, defined narrowly, is useful: it removes the execution burden of a disciplined rebalancing process and applies a consistent method without the behavioural drift that comes from doing it manually. As a category label applied broadly, it covers systems ranging from this kind of rules engine to tools making claims that no AI currently in public deployment can reliably support.

The current trends in quantitative investing reflect both: genuine progress in applying structured methods at the portfolio level, and the noise of products claiming more than the method can deliver. Separating one from the other is, increasingly, the work a careful investor has to do before the evaluation starts.

Acubic is a rules-based portfolio agent. It applies a published quantitative method, proposes trades for your review, and does not claim to know what the market will do next. If that description fits what you are looking for, the AI portfolio builder page explains where to begin.