BNY runs 100+ AI 'digital employees' with their own logins and human managers

What the agent does

Each digital employee works inside BNY's systems like a staff member would: it has an identity, access rights, and specific workflows. A human manager assigns its work and checks the output. Jobs include validating payment instructions and improving code security.

Why this is interesting

Most companies treat agents as features. BNY treats them as workers with identities. This makes access control and audit trails much simpler: you can see exactly which agent did what, and switch it off like any account.

What to copy

• Give each agent its own account with least-privilege access.
• Name a human owner for every agent.
• Log every action under the agent's identity.

Starter prompt

Written by us. A system prompt for a narrow operations agent.

You are {agent_name}, a digital operations worker. Your manager is {manager}.
Your only job: check incoming payment instructions against these rules:
{rules}

For each instruction, output PASS, FAIL (with the rule broken), or
REVIEW (with the reason). Never change an instruction yourself.
Every output must include the instruction ID.