Macro Is Changing. Can Agents Help Manage a More Dynamic Market?

^{This article is published by Figure Technology Solutions, Inc. (“FTS”). It is general information for institutional readers. It is not an offer to sell or a solicitation to buy any security, instrument, or service. It is not investment, legal, tax, or accounting advice, and it is not a recommendation of any venue, strategy, or allocation. FTS and its affiliates may develop, operate, or have commercial relationships with marketplaces, issuers, infrastructure, or allocation structures of the kinds discussed. Any allocation involves risk, including possible loss of principal. Past market conditions are not a reliable guide to future results. Fees, liquidity, settlement, and legal characteristics differ by instrument and venue and can change.}

The U.S. national debt crossed $40 trillion in August—faster than most forecasts anticipated—and has kept climbing through September. Treasury Secretary Scott Bessent answered by expanding buybacks of longer-dated Treasuries, first lifting the per-operation ceiling to at least $4 billion from early September and then running operations as large as $6 billion. Thirty-year yields have recently tested multi-decade highs; market reports dated September 24, 2026 described further pressure at the long end. The intervention has been framed as a liquidity measure, a signal that those yields do not fully reflect fundamentals, and a down payment on broader fiscal consolidation still to come. The long end has not cooperated. Yields remain even after the larger operations.

In a moment when the sovereign itself is actively managing the cost and composition of its own borrowing, idle cash on a corporate or fund balance sheet is no longer a passive line item. It is a live exposure—a balance whose opportunity cost moves with the yield surface and with the institution's own liquidity needs. On September 16, the Federal Open Market Committee raised the federal funds target by a quarter point to 3.75–4.00 percent, a unanimous vote and the first increase since 2023. Median projections now point to at least one further move this year. Short-term yields therefore offer more, not less, even as inflation remains elevated above the Committee’s 2 percent goal. The Committee cited elevated inflation and geopolitical uncertainty; energy and supply conditions have been part of the public debate, while economic activity is still expanding at a solid pace. Macro is changing across a variety of fronts in many conflicting ways, making the opportunity cost of unallocated balances volatile and dynamic with every basis-point move in the yield surface.

A treasury or fund without a deliberate strategy for those balances is leaving an opportunity cost that can be material when short-term yields and reachable instruments are changing.  Whether that gap is a policy issue depends on the mandate, liquidity schedule, risk limits, and governing documents of the specific institution. No single schedule or technology discharges that review.

That review has grown more difficult because the obligation has changed character. Cash management was long treated as an optimization exercise: sweep excess balances, ladder Treasuries, and chase a few extra basis points. Those tactics retain value, but they no longer suffice for mandates that require cash to be evaluated against a wider, faster-moving set of alternatives. Short-term rates reprice daily, liquidity needs evolve, and the available option set has become more accessible through programmatic innovation—private credit structures, securitized consumer credit, and other previously high-friction products. Each of those alternatives carries credit, liquidity, structural, and possibly prepayment or consumer-credit risk that many money-market instruments do not. Reviewing idle cash management only quarterly may not discharge the duty on a sufficient schedule and may lag the market for certain mandates. Whether a more frequent cycle is warranted depends on the facts and circumstances and is not a universal rule.

A continuous market does not, by itself, impose a duty of continuous reallocation. It does, however, raise the cost of a stale cash policy when the mandate contemplates yield-seeking within stated limits.

Continuous reallocation across a growing marketplace exceeds what most desks can do line by line, every day. A desk can monitor a limited set of instruments and rebalance periodically. It cannot evaluate, compare, and move capital in real time across dozens of venues whose terms, liquidity profiles, and risk characteristics shift constantly. Dashboards that surface recommendations do not close the gap, because capital still has to move. Modeling without execution leaves an operational gap between analysis and authorized action.

Taken seriously, some institutions may consider delegated execution: agent-assisted trading to complement agent-assisted modeling, only inside an authorized mandate, with human accountability remaining in place. Once an agent can observe the surface, reason within policy, and execute, the operational bottleneck narrows. That capability is still limited, however, with model error, unauthorized action, incomplete data, and failed settlement remaining material risks. 

Most institutions cannot currently make that hand-off. Point an agent at a corporate treasury or fund balance sheet and the model reasons fluently about rates, duration, credit, and liquidity. But ask it to act, and it encounters any number of impediments. Asset terms residing in PDFs. Disjointed performance data arriving monthly by email or by portal. Settlement remaining T+2 and human-gated. Counterparty onboarding, wire instructions, and approval chains sitting outside machine-readable interfaces. The constraint is not intelligence but an interfaceable surface for execution.

Agents cannot be bolted onto financial infrastructure designed for humans and be expected to operate reliably. Settlement, custody, data, and constraint layers must be constructed so the agent works inside them. Four mechanisms are typically discussed as design conditions, albeit not guarantees of performance:

These features describe an operating environment in which software can initiate actions that humans can review, constrain, and audit. They do not necessarily make trading successful, nor do they remove the need for oversight. Delegation without restrictions is inconsistent with ordinary control and would be an abdication of responsibility. 

That said, an agent that is properly sandboxed—whose every action is bound by machine-enforceable rules and fully logged—can produce a more complete audit trail than a desk that records judgment in only email and notes. While the agent may not automatically be safer, more accurate or more suitable for the task assigned, properly designed agents may ultimately ensure better controls. 

Defined and encoded constraints become another operational burden. For firms that want software to act inside policy at higher frequency, that burden is increasingly a design choice. The question is whether the firm needs an operating environment that can keep pace with a larger, faster-changing set of alternatives—not whether any particular agent, venue, or schedule will do so. (Accuracy and speed are not assured. Yield opportunities differ in risk, liquidity, cost, and tax treatment.) 

Measured against these four mechanisms, common cash and near-cash venues differ. 

Traditional bank sweeps and many money-market funds are often slower to expose programmatic subscription, redemption, and constraint interfaces. They may offer same-day or next-day liquidity, stable or near-stable principal targets, and, in some deposit structures, deposit insurance up to applicable limits. Those features can often matter more than machine-accessibility for a cash mandate.

Private credit and many securitized products, including securitized consumer credit, often lack machine-readable terms and real-time performance feeds. They may offer higher contractual yield in exchange for credit risk, structural complexity, longer or uncertain liquidity, and principal fluctuation. (Not cash equivalents). 

The intention of onchain and programmatic private-market rails is often to enable settlement, position state, terms, and constraints that can be read by software. That design can reduce certain interface frictions. It’s also important to note that this does not make those rails superior for every mandate. Material limitations must also be considered, including smart-contract and key-management risk, venue and counterparty risk, liquidity that can gap or halt, principal fluctuation, limited or no deposit insurance, evolving regulation, fees, and irreversible settlement errors.

Despite this, we are beginning to see the proliferation of more onchain enabled products. In fact, as BlackRock’s September 2026 paper, The Machine-Native Economy, argues agentic commerce and machine-to-machine payments may increase demand for blockchains and other programmable payment infrastructure, as some traditional financial rails serve automated software transactions less efficiently. That is a research thesis, however, rather than a requirement that any allocator use onchain assets—BlackRock also disclaims that much of the activity described still remains nascent.

Regardless, those rails are becoming more prominent. Two-sided marketplaces that render traditionally illiquid, passive-yielding products accessible have already compressed onboarding, settlement, and data into programmable form, while also making these asset classes more available than ever before to a wide swath of investors. (Greater availability is not the same as greater suitability. Instruments that were illiquid remain subject to liquidity windows, gates, or thin secondary markets unless the specific venue terms say otherwise.)

In these environments, settlement, position state, and constraints can be built so an authorized agent can operate with fewer manual gates, even as most institutional setups maintain custody approval chains that still insert people. 

The growing breadth of financial instruments available to this type of automation increases operational complexity. While increased complexity does not make constrained automation necessary, it does make agentic allocation one practical response some firms are testing as they attempt to keep pace with the ever changing available set of opportunities in the market.

Agentic trading does not simply accelerate what a skilled desk already performs. It reduces the need for every treasury or fund to staff and maintain that desk. Continuous evaluation across a multi-venue surface no longer requires a dedicated quantitative team, even under tight constraints. The institution can allocate into an agentic strategy whose mandate, liquidity rules, and risk limits are already encoded. Curator-style structures become the delivery vehicle, with policy set once, and an agent designed to operate inside it.

Funds and onchain traders are beginning to start experimenting with this approach. Some are already deploying their first agentic vault strategies.

The model is evolving quickly, but open questions remain. Custody approval chains still require human gates in many institutional setups. Standardized audit trails that satisfy risk committees and regulators are incomplete. Model risk—validation, versioning, and override of an agent’s reasoning—still requires further design. Regulatory comfort with scaled delegated execution is still forming.

None of this makes delegated execution mandatory, or even easy. It does mean that as accessibility expands for both agents and humans, and as the opportunity cost of idle cash continues to move with the yield surface, more allocators will have to decide whether their current cash process can keep up with a wider, faster set of alternatives.

Under yield-seeking mandates, unallocated capital is often expected to be reviewed as the opportunity set moves. Most of it still cannot be—at least not efficiently. Agents may be used to compare instruments and, where authorized, initiate reallocation inside encoded policy, while they remain rudimentary in their ability to run the book. 

The rails they would use are already being built, however; venues where settlement, positions, and constraints are becoming faster and more programmable. Institutions can learn those marketplaces, issuers, and infrastructure firms now, including differences in cost, liquidity, safety, insurance, and tax treatment.

The agentic economy is still forming. As the infrastructure it may run on is being built, institutions can prepare by asking themselves a few questions:

Familiarity with the new opportunities becoming available—through study or allocation—as well as a good grasp on current state, is what may help  set your organization up for success if the proliferation of AI agents makes delegated execution more operationally viable. 

Important Disclosures

FTS and its affiliates, including Figure Markets, Figure Markets Credit, Figure Payments, Figure Securities, Figure Equity Solutions, and related entities, develop, operate, or may have commercial relationships with marketplaces, issuers, infrastructure, protocols, or allocation structures of the kinds discussed in this article. Those relationships may include onchain and programmatic rails, tokenized or yield-bearing cash alternatives, private credit and securitized consumer-credit products, lending and borrowing marketplaces, transfer-agency or registry services, and related software. FTS and its affiliates may receive fees, spreads, token economics, or other economic benefits from those activities. Readers should assume FTS has a financial interest in the broader adoption of programmable market infrastructure.

Nothing in this article creates or describes a duty owed by FTS to any reader. Whether unallocated cash must be reviewed, how often it must be reviewed, and whether delegated or agent-assisted execution is appropriate are facts-and-circumstances questions that depend on each institution’s mandate, governing documents, liquidity schedule, risk limits, and applicable law. A continuous market does not, by itself, create a continuous reallocation duty.

References to bank sweeps, money-market funds, Treasuries, private credit, securitized consumer credit, onchain assets, stablecoins, tokenized instruments, vaults, or “agentic” strategies are illustrative. They are not cash equivalents unless the specific instrument’s terms and applicable regulation provide. Higher contractual yield generally involves credit, liquidity, structural, prepayment, consumer-credit, smart-contract, key-management, venue, counterparty, operational, and regulatory risk that many money-market instruments do not. Encoded constraints may mitigate specified policy breaches; they do not mitigate market loss, liquidity stress, model error, failed or irreversible settlement, or custody failure.

Discussion of artificial intelligence, agents, or delegated execution is conceptual. Models can be wrong. Data can be stale or incomplete. Sandboxes and override controls can fail. Logs improve reconstructability; they do not validate a model or make an agent safer or more suitable than a human desk. Many institutional setups still require human custody, payment, or compliance approval. Regulatory treatment of scaled delegated execution remains unsettled.

Citations to third-party research, including BlackRock’s September 2026 paper The Machine-Native Economy, are provided for context. Those publications are not affiliated with FTS, do not endorse FTS products, and do not require any allocator to use onchain assets, agents, or any particular venue. Third-party views are their own and may be incomplete or outdated.

Forward-looking statements—including statements about rates, fiscal policy, marketplace development, agent capabilities, and “what next”—reflect conditions as of September 2026 and are inherently uncertain. FTS undertakes no duty to update this article.

Readers should perform their own due diligence and consult their own legal, compliance, tax, and investment advisors before changing cash policy, adopting software agents, or allocating to any instrument or venue mentioned here.