Field Notes
Practical notes on AI agent deployment.
Short essays on where agents fit, how teams should sequence implementation, and which technical choices actually change the operating model.
Sanctions Screening Agents for Order and Payment Holds
A sanctions screening agent should compare counterparties to current restricted-party lists, assemble a match file, and hold the order or payment until a named owner decides. It should not clear a hit, block property, or file an OFAC report on its own.
Certificate of Insurance Tracking Agents for Vendor Compliance
A certificate of insurance tracking agent should compare each certificate to the contract's insurance requirements, name every gap, and route the decision to a risk owner. It should not accept coverage, request endorsements, or release payments on its own.
Chargeback Representment Packets for Payments Operations
A chargeback representment agent should assemble evidence against the cardholder claim and route accept-or-counter decisions. It should not submit to the issuer on its own.
Access Recertification Agents for Periodic User Reviews
An access recertification agent should assemble evidence packets and route keep-or-revoke decisions, not remove entitlements on its own.
Shadow AI Inventory for Unsanctioned Agent Discovery
A shadow AI inventory classifies unsanctioned models, custom GPTs, and embedded SaaS agents, then routes keep, govern, or retire decisions. Discovery without an owner is just a longer list.
Privacy Request Packets for DSAR and Consumer Rights
A privacy-request agent should verify identity, locate systems, and assemble an export or denial packet. It should not release a copy or delete records on its own.
Agent Incident Rollback for System-of-Record Repairs
Agent rollback is a compensating repair in the system of record. Replaying a checkpoint does not unpost overtime, unsend a notice, or restore a closed period.
Agent Cost and Latency SLOs for Production Queues
An agent SLO should bind cycle time and cost per completed case to a named queue. Token charts without an error budget do not tell operators when to shed load.
Third-Party AI Vendor Risk Reviews
An inbound AI vendor review should produce a use-case packet covering models, data flow, subprocessors, and change notice. A recycled SOC 2 questionnaire is not enough.
Citation Freshness for Retrieval-Backed Agents
A retrieval-backed agent should cite a current controlled source or stop. Filename annotations are not enough when the work instruction has a revision stamp.