Capital. Connected intelligence.

Intelligence at the centre of the institution

USD2’s operating thesis brings USDT capital, agricultural banking and physical trade together with centralized AI intelligence. The fund and WAB are being designed around an internal system that coordinates their daily work: collecting evidence, maintaining context, preparing decisions and following each obligation through to completion.

The ambition is a small permanent team with broad operating capacity. People contribute relationships, reliable data, prompts, commercial judgment and regulatory engagement. The platform connects that contribution to repeatable work across finance, service and oversight. Fund 1 remains a 150 million USDT target in formation, and WAB remains a bank in formation.

Own the models and the operating platform

USD2 and WAB will develop proprietary AI models alongside their operating platform. The first model families are centred on institutional work: understanding documents, matching entities, interpreting trade milestones, supporting credit analysis, identifying exceptions and preparing reporting. Their quality depends on permitted data, domain expertise and representative evaluation.

Model development can include task-specific training and adaptation of appropriately licensed architectures. The development plan must establish ownership of weights and software, data rights, reproducibility and the cost of serving each task. A model’s usefulness is demonstrated through its workflow results.

The platform is for internal operation. Its purpose is to improve the fund and bank’s execution; external software licensing is outside the current business model.

The human network supplies the real-world context

Business development brings in a farmer association, cooperative, supplier, buyer or government relationship. Subject-matter specialists turn documents and local conditions into verified inputs. People responsible for policy and compliance translate obligations into maintained operating rules. Customer-service teams resolve the cases that need a conversation or local understanding.

Prompts are a working interface: ask for an analysis, supply context, correct an error or develop a workflow. A prompt that changes a material policy goes through its approval process before it changes execution permissions. Knowledge and authority have separate records.

One workflow from evidence to action

A supplier invoice can be connected to its contract, delivery milestone, approved beneficiary and facility limit. The intelligence layer prepares a structured decision packet and identifies missing or conflicting evidence. Defined controls evaluate the instruction before the authorized execution system acts.

The same approach applies to customer onboarding, seasonal-credit preparation, treasury reconciliation and LP reporting. Each workflow retains its sources, analysis, permissions, approval, execution receipt and subsequent reconciliation.

1. 01

   Receive and verify

   Classify the entity, documents and data rights. Resolve missing or inconsistent evidence.

2. 02

   Analyze and coordinate

   Use the appropriate proprietary models, verified calculations and institutional context.

3. 03

   Authorize and execute

   Apply policy limits, scoped tools and the human approvals required for the action.

4. 04

   Reconcile and learn

   Compare the outcome with the instruction, resolve exceptions and improve the evaluated workflow.

Central coordination, distinct institutional authority

USD2’s investment and treasury records remain separate from WAB’s customer accounts, credit decisions and regulated banking activity. Access between those environments requires an approved purpose and permission. A common coordination platform does not combine their money, ledgers or legal responsibilities.

AI can prepare and monitor a transaction. Payment destinations, limits, signatures and custody rights are enforced through controls outside the generative model. Accountable people retain material investment, credit, policy and regulatory decisions, with an independent route to challenge the work.

AMCM’s technology guideline expressly addresses AI governance, risk assessment, records, data protection and fallback. WAB’s implementation will be developed within its licensing and supervisory process. The diagram and workflow are an operating design, rather than a live autonomous banking service.

Measure capacity and complete cost

A useful economic test compares the complete cost of a verified outcome with its previous operating baseline. The measure includes model development, inference, data preparation, human review, security, support and rework. Token usage or a low payroll alone cannot establish efficient banking.

The objective is to increase verified work completed per responsible person while preserving service quality, credit discipline and control. A more capable team may choose to serve more customers and trades rather than reduce its existing workforce. Staffing follows the institution’s responsibilities and actual workload.

The next chapter is operational

The September 2026 a16z State of Markets report describes expanding AI use, company-reported efficiency and a gap between adoption and sustained measurement. It also shows that higher AI intensity can accompany more technical hiring. These observations support testing a lean operating model; they do not establish USD2’s savings or WAB’s staffing needs.

The advantage USD2 aims to build is cumulative institutional knowledge: verified records, evaluated models, completed workflows and relationships that improve origination and service over time. The report informs that direction while the business must establish its own evidence.

Build capability in stages

The initial stage maps the fund and bank’s work, defines data rights and creates evaluation cases. Read-only analysis and draft preparation establish a baseline before bounded automation is introduced. Each material expansion of permissions follows performance, security and control review.

The shareholder room sets out the proprietary model portfolio, human responsibilities, workflow authority and delivery sequence. Technology investment will have its own approved budget and ownership arrangements; it is not an additional allocation assumed within Fund 1’s capital target.