The Autonomous Treasury: How AI and Programmable Money Are Rewriting Corporate Finance
The Autonomous Treasury: How AI and Programmable Money Are Rewriting Corporate Finance
April 2026 — Corporate treasury is undergoing a structural transformation that moves beyond incremental digitization. Three converging technology vectors—agentic artificial intelligence, programmable payment rails, and tokenized deposits—are enabling a new operational paradigm: the autonomous treasury. Unlike earlier phases that focused on visibility and dashboards, the current cycle emphasizes self-executing cash operations where machines initiate, validate, and settle transactions without human intervention.
This article examines the architectural logic, real-world deployments, governance constraints, and organizational bottlenecks that define this transition. The evidence suggests the primary limiting factor is not technological capability but institutional readiness and regulatory adaptation.
---
The Execution Gap: Why Visibility Without Action Is No Longer Enough
“Treasurers don’t lack visibility anymore; they lack widgets that can act on that visibility in real time,” stated Sayantan Chakraborty, head of Digital Payments at Fiserv (Source 1: Primary Interview, Global Finance Magazine). This observation captures the central tension in modern treasury operations: legacy systems provide observational dashboards that display cash positions, forecast deviations, and risk exposures, yet they do not execute corrective actions.
The autonomous treasury model addresses this gap through three investment priorities:
1. Real-time cash positioning across all bank accounts and legal entities, updated continuously rather than at end-of-day batch cycles.
2. Rule-based, just-in-time money movement across multiple payment rails, where funds transfer automatically when predefined thresholds trigger.
3. Tokenized deposits and programmable payments that embed settlement logic directly into the payment instrument, eliminating manual reconciliation steps.
The distinction from prior automation phases is qualitative. Traditional treasury management systems (TMS) automated reporting. Current systems automate decision execution. The difference is between a dashboard that shows a projected shortfall and a system that borrows from the optimal source before the shortfall materializes.
---
Siemens and the Programmable Payment Milestone
The most documented production deployment of autonomous treasury technology occurred in late 2023, when Siemens adopted J.P. Morgan’s programmable payment feature on its Kinexys platform (formerly Onyx), using JPM Coin as the settlement asset (Source 2: J.P. Morgan Product Documentation; Siemens Cash Management Disclosure).
The mechanism is architecturally significant. Siemens implemented a just-in-time funding structure where funds transfer only when a payment obligation becomes due. The system monitors cash balances across Siemens’ global account structure and autonomously sweeps funds from a central cash pool when an individual account balance drops below a configurable threshold. This eliminates the need for treasury staff to manually initiate intercompany loans, manage liquidity buffers, or predict intraday cash requirements.
Heiko Nix, global head of Cash Management and Payments at Siemens, identified the critical success factor: “In my experience, the biggest challenge is not technology, but the mindset shift in finance and treasury. For almost every technical problem, there is a solution. But simplifying entrenched processes and changing how people think about treasury and its role takes significantly more time and effort” (Source 3: Primary Interview, Global Finance Magazine).
The Siemens case demonstrates that programmable payments are not theoretical. JPM Coin, operating on a permissioned blockchain, enables atomic settlement—payment instruction and value transfer occur simultaneously, removing the settlement risk inherent in traditional wire transfers. The autonomous sweep logic operates within a governed framework: policy rules define maximum sweep amounts, counterparty limits, and regulatory compliance parameters. Human oversight remains, but shifts from transaction approval to exception monitoring and rule tuning.
---
Agentic AI in Action: HighRadius’ 186 Agents and the 90% Touchless Target
In February 2026, HighRadius released 186 agentic AI agents designed for treasury and working capital operations (Source 4: HighRadius Product Release Notes, February 2026). The scale of this deployment signals a shift from AI as a forecasting tool to AI as an autonomous execution layer.
HighRadius reported that cash application—the process of matching incoming payments to outstanding invoices—and cash forecasting have already achieved 90% touchless automation. The remaining 10% involves edge cases requiring human judgment, such as disputed invoices or incomplete remittance data.
Sashi Narahari, CEO of HighRadius, stated a target of 90%+ end-to-end touchless automation by 2027 (Source 5: Primary Interview, Global Finance Magazine). “Agentic AI will push treasury from once-a-day instructions to continuous, just-in-time funding: as soon as execution matches intent across all rails,” Narahari explained.
The architecture of these agents differs from earlier robotic process automation (RPA) deployments. RPA systems follow rigid scripts and break when input formats change. Agentic AI systems can interpret unstructured data, adapt to process variations, and execute multi-step workflows across disparate systems—ERP, bank portals, TMS, and reconciliation platforms—without requiring API integration for every link in the chain.
For working capital management, the implications are structural. As Narahari noted, “AI can transform working capital management from a retrospective reporting function into a forward-looking control tower. Instead of focusing on past events, you can optimize for the future in real time” (Source 5). This shifts the treasurer’s function from historian to strategist, but only if organizations restructure teams accordingly.
---
Governance as the Gatekeeper: Kyriba’s Claude Deployment and the Regulated Environment
Kyriba’s approach to autonomous treasury, branded as TAI (Treasury AI), uses Anthropic’s Claude as the underlying large language model within a governed deployment framework (Source 6: Kyriba Product Architecture Disclosure, 2025). The implementation is instructive because it addresses the regulatory constraints that define treasury operations.
John Stevens, senior vice president and global head of Capital Markets, Financial Institutions & Working Capital at Kyriba, emphasized the deployment philosophy: “Think of it as an AI-powered autopilot added to an older cockpit. Policies are enforced, actions are executed, and audit trails are preserved without forcing a full-core replacement on day one, under the watchful eyes of a trained cockpit and cabin crew” (Source 7: Primary Interview, Global Finance Magazine).
The governance architecture includes:
- Strict data access limits: The AI model cannot access all corporate data; it operates within a bounded data environment defined by role-based permissions.
- Complete audit trails: Every AI-generated action is logged with the model’s reasoning, the data inputs used, and the human approver (if required).
- Human-in-the-loop validation: Critical actions—large wire transfers, counterparty changes, debt covenant compliance—require human authorization, while routine sweeps and rebalancing execute autonomously.
The decision to use an external model (Claude) rather than a proprietary system introduces sovereignty concerns. Gilly Wright, whose role involves treasury technology strategy, stated: “Some organizations will require sovereign options for policy or jurisdiction reasons, but most regulated treasuries are looking for governed AI: strong models, used in a way that is secure, auditable, and designed for real operational control” (Source 8: Primary Interview, Global Finance Magazine).
This distinction is critical. The autonomous treasury does not eliminate human governance; it redefines it. Instead of approving individual transactions, treasury staff validate policy logic, monitor exception reports, and audit model performance. The bottleneck shifts from manual processing capacity to rule-set quality and anomaly detection capability.
---
The Organizational Readiness Gap
Despite proven technology, adoption of autonomous treasury remains limited to early adopters. The constraint is not technical but organizational. Three factors determine readiness:
First, process standardization. Autonomous execution requires deterministic rules. Organizations with fragmented payment processes, inconsistent data formats, or manual exception handling cannot simply layer AI on top; they must first rationalize their operational baseline.
Second, trust calibration. Finance teams accustomed to verifying every transaction must transition to a monitoring and exception-handling mindset. This requires retraining and performance metrics that measure system reliability rather than individual transaction accuracy.
Third, regulatory alignment. As John Stevens noted, “We don’t see a single out-of-the-box ‘autonomous’ product replacing the diversity of treasury needs” (Source 7). Regulatory requirements vary by jurisdiction, industry, and transaction type. A global treasury operating across 50 countries cannot deploy a uniform autonomous system; it must configure jurisdiction-specific rule sets, each with its own compliance framework.
The Siemens experience confirms this. The autonomous cash management deployment succeeded because Siemens first simplified its global account structure, standardized cash pooling agreements, and established clear governance for exception handling. Technology was the final layer, not the starting point.
---
Future Trajectory and Market Predictions
The autonomous treasury market will likely evolve along three dimensions through 2028:
1. Deep integration with central bank digital currencies and tokenized deposits. As programmable money becomes available on regulated ledgers, autonomous systems can settle transactions atomically rather than waiting for batch-clearing cycles. This reduces intraday liquidity requirements and eliminates settlement risk.
2. Cross-platform agent interoperability. Current agentic AI deployments are siloed within single vendor ecosystems. Future systems will require agents that can negotiate payment terms, optimize funding sources across bank relationships, and execute hedging strategies across multiple counterparties—all without human intervention.
3. Regulatory sandboxes for governed AI. Regulators in major financial centers are developing frameworks for autonomous financial operations. Expect structured approval processes for AI-driven treasury functions, similar to the algorithmic trading approvals in capital markets. Firms that engage early with regulators will have competitive advantages.
The technology is ready. The question is whether corporate finance organizations can reorganize their processes, retrain their people, and renegotiate their trust relationships with machines fast enough to capture the efficiency gains. The autonomous treasury does not eliminate the treasurer. It elevates the role—but only for those willing to surrender direct control to systems they cannot fully predict.