Varick's Agents clear every exception queue before the cutoff.

With one queue that runs against a clock, often AML alerts, wire repairs, or KYC refresh. In a scoped 4-week pilot, our forward-deployed engineers mine the case and payment logs with read-only access, interview the analysts, and set the baseline the agents are measured against.

Under the model risk and AI governance policy your validators already apply. SR 26-2, which replaced SR 11-7 in April 2026, leaves agentic AI outside its scope, so your own policy sets the bar. Each agent is benchmarked on your own historical cases before go-live, runs in shadow mode on live work first, and logs every input, rule, and output so validation can test it.

The NAIC's 2023 model bulletin on insurers' use of AI, since adopted by more than 20 states, asks for a written governance program, testing, and documentation. Every agent ships with its benchmark results, its rules, and a log of each decision it supported, which your AI governance committee can file against that program.

No. Agents set up claims, prepare submission files, and draft coverage positions. Underwriters set every rate and bind within their own authority, and adjusters decide coverage, reserves, denials, and payments. Routine endorsements issue only inside the authority table your underwriters set, and every other change goes to an underwriter.

Yes. Agents read the signed investment management agreement, its amendments, and each side letter, and link every coded rule or fee term to the clause it came from. Compliance approves any change before it goes live in pre-trade or post-trade testing.

No. Agents run in your cloud under identities your team provisions, with read-only access to start. Customer, account, and transaction data stays inside your boundary and is never used to train a model.