Agentic AI · Regulated financial services

Shipping AI agents where mistakes are expensive.

I lead agentic AI product work in a bank. This site shows how: the patterns, the decisions, and the tradeoffs behind them. The scripted run here is one of them: make the call when it asks you.

Read the Field NotesSee the work
Illustrative run · scripted example

A corporate client asks to add a user to their payments portal. The agents do the work; a person still decides what gets sent.

    Activity · from the GitHub API

    I build in bursts

    A problem worth solving gets a few intense weeks of evenings and weekends with Claude Code, then the day job takes the front seat again. Every commit across my repos over the last year, one square per day.

    Commits in the last year
    104
    Weeks with commits
    12 of 52
    Longest run
    2 weeks
    Last commit
    Oct 23
    Last 6 months

    Field Notes

    Patterns for shipping AI agents where mistakes are expensive. Each one runs as an interactive example.

    01
    Human always sends

    Agents draft and fill forms. A person owns every message that leaves the building.

    02
    Confidence gating

    Let the model act only where it has earned it. Everything else gets a head start, not an autopilot.

    03
    Evals before scale

    Decide what "good enough" means, measure it on real cases, and write the release rule down before the results come in.

    04
    Staged release gates

    Match the controls to the risk. Let low-risk changes flow, and make high-risk ones earn their way out in stages.

    05
    Orchestrators, supervisors, and MCP

    Split planning from doing, give every agent the narrowest tools that work, and have something watching what they actually try.

    All notes →

    The Work

    Decision write-ups: the constraints, the options on the table, and why I chose what I chose

    PilotFinancial services · Recent

    Bringing AI agents into client servicing

    How I approach putting agents in front of client-servicing work: agents read, route and draft, and a person sends every message.

    Method write-up · results are internal
    In progressJPMorgan Chase · 2024 - Present

    A GenAI rules system that knows when to ask

    Strategy for a GenAI system that finds duplicate and conflicting decision logic across thousands of product configurations, with confidence-based routing between automation and expert review.

    3,000+
    Configurations in scope
    40%
    Less manual maintenance (target, not yet a result)
    ShippedJPMorgan Chase · 2021 - Present

    Self-service product data, with controls that pass audit

    Scaled a self-service product data platform from 0 to 400+ users, replacing $50K, six-month engineering projects with same-day changes under a risk-based change framework.

    0 → 400+
    Users
    1,200+
    Product changes a year
    ShippedJPMorgan Chase · 2022 - 2023

    Fixing client onboarding at the point of selection

    Cut onboarding rejections by 80% with eligibility rules that block invalid product combinations up front, and cut onboarding time by 65% by modernizing the APIs behind it.

    25% → <5%
    Rejection rate (within the reviewed product subset)
    3,500+
    Rejected applications avoided a year
    All case studies

    Writing

    Writing on product, AI, and building at the intersection of both

    View all articles