Quant trader automation has two clocks: execution and risk ownership

We expect bounded trading workflows to automate faster than the responsibility for deciding how much capital a strategy deserves. A trader can delegate repeated decisions inside a mandate while retaining the authority to change that mandate. Keeping those two clocks separate makes the automation forecast both stronger and more useful.

Jane Street's quantitative trader role combines signals, strategy execution, modeling, risk management and business development. A forecast about that role must address all of those activities. Our 2026-2028, 2028-2031 and 2031-plus windows are DXRG scenario assumptions, rather than company announcements or estimates published by a benchmark organization.

Imagine a desk with a frozen strategy and an explicit inventory ceiling. Orders arrive, partial fills change the account, and an external feed stalls. Much of the repeated workflow can be delegated if authoritative state is available and execution controls enforce the ceiling. The consequential decision arrives when a trader asks whether the ceiling should change because market conditions have become unusual.

Execution within a mandate, 2026-2028

In our near-term scenario, agents increasingly perform preparation and monitoring around existing execution systems. They investigate anomalies, summarize inventory, compare permitted actions and reconcile outcomes. For liquid markets with accessible interfaces, selected agents can also propose actions inside tightly specified limits. This does not require an LLM to replace the fast path of every trading system.

We define mid-horizon decisions here as positions evaluated over hours to days. Their information-processing demands differ from latency-sensitive market making. An agent that interprets a filing over several minutes could be useful for the former and entirely unsuitable for the latter. The opportunity depends on the decision deadline, available information and cost of a mistaken action.

The main adoption requirements are fresh account state, recoverable order submission and controls independent of the proposal generator. When an acknowledgment is missing, the agent must resolve the existing order before creating another. When market data is stale, the runtime needs an explicit permitted response. A fluent explanation has little value if the account silently accumulates unintended exposure.

Our confidence in this phase is relatively high for repeatable desk support and lower for unsupervised action during unfamiliar incidents. Measure escalation frequency and unresolved exposures alongside completion. A workflow that appears autonomous because a person repairs it every afternoon has transferred the work to a less visible location.

Risk allocation across mandates, 2028-2031

The next scenario moves from individual actions to bounded allocation decisions across strategies. An agent could recommend smaller exposure after a change in liquidity, investigate conflicting signals or suspend a candidate within previously approved rules. The mechanism requires joined portfolio state and an evaluation of how strategies fail together.

The current METR horizon dashboard defines horizons by human expert task duration and a selected success probability. A 50% horizon means a modeled success rate of one half, which is far below the reliability required for many capital responsibilities. Software-task progress supports the possibility of longer autonomous investigations. Financial adoption additionally requires evidence about stress losses and recovery.

An illustrative evaluation would replay the same delayed feeds and changing inventory through human-led and agent-led workflows. Freeze the risk mandate, compare completion before the decision deadline and measure breaches, duplicate actions and capital tied up in unresolved state. Separately assess whether the allocation recommendations improve outcomes on held-out periods. This prevents operational reliability from being mislabeled market skill.

Our dates move earlier if agent allocation remains dependable under correlated shocks and independently reviewed incidents. They move later if a strategy looks good alone but fails alongside the rest of the portfolio. A trader's practical knowledge often concerns those interactions.

Ownership remains a different decision

From 2031 onward, broader delegated desk operation is plausible in our scenario for well-instrumented markets. Ultimate capital ownership still involves deciding tolerable losses, financing inventory and accepting business obligations. Technical capability can make that ownership cheaper to exercise while the authority stays with a firm or account owner.

Established firms can benefit from the same automation. Their capital, low-latency infrastructure and market relationships can complement agents rather than disappear when software becomes cheaper. We expect routine execution support to become widely accessible sooner than those advantages become replicable.

The profession's center of gravity may move toward designing mandates, validating allocation and resolving unusual market events. The decisive receipt is a sequence of successfully handled situations with measured economic consequences. Counting automated order proposals answers a much smaller question.