Agentic AI · Systems · Automation

Art
Tiongson

Operations Builder · Directs Agentic AI |

I direct AI agents to build production software. My domain is operations and accounting: I own the problem end to end, requirements, architecture, and source-verified validation, while agents write the implementation under my standards.

Art Tiongson
Scroll
About

How I
Got Here

Career Snapshot
$65M
Operations managed
2-3 days
1 click
Monthly reconciliation time
6
Tools shipped
5
Languages
AI
Agent-directed

I ran $65 million in store operations before I ever wrote a line of code. That's why I build differently, and why I now direct AI agents to ship the systems I used to wish for.

Before moving into product accounting, I led store operations at a major national retailer, managing a $65M book of business, leading cross-functional teams, redesigning operational workflows, training staff on new systems, and owning the P&L outcomes.

I know what operational friction looks like because I've spent years fighting it from the inside.

When I moved into product accounting, I found myself working inside Microsoft Dynamics NAV, running monthly close processes and managing inventory reconciliation. Within my first month, I looked at a reconciliation workflow that consumed 2-3 days every cycle and I decided to fix it.

I fixed it. I directed AI to build an automated Power Query pipeline pulling live data through ODBC connections directly from our ERP database. What took days now takes one click. The real lesson was not that AI could write code for me. It was that I could direct it to build real systems, and that the constraint was my judgment about what to build, not the typing.

Monthly Close Reconciliation
Before: 2-3 Days
1 Compile hard goods inventory
2 Manually create Excel columns
3 Write VLOOKUP formulas by hand
4 Reconcile data line by line
5 Review numbers by site
After: Seconds
One-Click Refresh
Power Query + ODBC → live ERP data, auto-reconciled

So I escalated the same skill: from automating one report, to directing agentic coding workflows in Claude Code, to running multiple AI agents in parallel under my direction and building the custom tooling (MCP servers) that extends what they can do. The output: Jarvis, my private, always-on AI assistant with a governed agent runtime, full-duplex voice, two-way iMessage, and a human-gated self-modification loop; an open-source Excel MCP server that gives agents token-bounded access to spreadsheets; Ombri and ArtFitness on iOS; Music Tool on iPad; an internal inventory portal in ASP.NET Core; and this site. I directed agents to build all of it.

I don't position myself as a traditional developer. I'm an operations builder who directs agentic AI to ship at the speed of a full engineering team, and who builds the agent tooling itself. I see where processes break down, and I have the judgment and the tooling to fix them.

Operating Model

How I Work

I don't write the code, and I never have. I direct AI agents to build it. My job is the part the agents can't do: owning the problem, setting the architecture, holding the standards, and verifying the result against the source. I work the way a lead does, except my team is a fleet of AI coding agents running in parallel.

01 Own
I own the problem end to end
Requirements, architecture, data decisions, and the definition of done are mine. I come from operations and accounting, so I read process bottlenecks fast and know what a correct result actually looks like before a single line is written.
02 Direct
I direct agents, in parallel
I orchestrate multiple AI coding agents at once, each in its own lane, coordinated through written handoff docs and strict version control. When two agents once built the same component independently, I caught the collision in git, backed out the duplicate, and kept the better implementation.
03 Govern
I set the standards they work within
The agents operate inside rules I define: deny-by-default permissions, propose-don't-apply self-modification, and test-before-promote. I also build the custom tooling (MCP servers) that extends what they can safely do.
04 Verify
I verify against the source
Agents are fast, not infallible. I validate every output against the real data. That discipline is how I caught an accounting process overstating a material line by ~57% and surfaced redundant spend others had missed.
Proof

Impact

$65M
Operations managed
Book of business owned in retail store operations before moving into product accounting.
~57%
Costing error caught
A bug in the prior process was overstating a material line. Found by validating agent output against source data.
~$40K/yr
Redundant spend cut
Reverse-engineered a legacy costing database and surfaced software spend used by only two people.
Days → 1
Monthly reconciliation
A multi-day manual close became a single refresh via a Power Query and ODBC pipeline against the ERP.
Skills

What I Work With

AI / Agentic
Claude API & Agent SDK Claude Code Model Context Protocol (MCP) Agentic Dev Workflows Multi-Agent Orchestration
Operations & Finance
Product Accounting Inventory Reconciliation Monthly Close P&L Management Process Optimization Operations Leadership
Technical
HTML / CSS / JS C# / ASP.NET Core Python SQL Power Query / M ODBC
Tools & Platforms
Microsoft NAV / ERP Excel Automation Git / GitHub VS Code Discord API Claude / ChatGPT
Selected Work

Things I've Built

Self-Hosted AI Platform Always on

Jarvis

A self-hosted AI assistant platform with the Claude API as its reasoning core. Full-duplex voice, two-way iMessage, and smart-home control, with persistent memory and a permission layer that allow-lists and audits every tool call.

Claude Agent SDK Python Voice Home Assistant
iOS Application TestFlight soon

ArtFitness

A native iOS workout app with an AI coaching engine that gives real-time, in-workout feedback informed by your training history. A REST API exposes workout data to other tools, and response caching keeps per-user inference cost in check.

SwiftUI SwiftData Claude API REST API
Open Source · MCP Server Open source

Excel MCP Server

An open-source MCP server that lets an AI agent read and manipulate Excel workbooks directly, with token guardrails that keep large-file operations inside the model's context window.

Python Model Context Protocol Excel
Archive

Old Projects

iOS Application In progress

Ombri

Finance by the hour: a personal finance iOS app with a time-of-day aesthetic. Morning, afternoon, sunset, and night themes shift the interface as the day moves. Currently in beta.

Swift SwiftUI iOS PWA Preview
iPadOS Application Side project

Music Tool

An iPad reference tool for guitar, ukulele, and bass. Chord library with color-coded voicings rendered on a live fretboard, scale visualization, chromatic tuner with autocorrelation pitch detection, and a songs workspace with chord diagrams.

SwiftUI iPadOS 18+ AVAudioEngine Canvas
Internal Tool

Inventory Management Portal

Replaced slow legacy system lookups with an instant, cached dashboard, giving the team real-time visibility into inventory issues.

ASP.NET Core C# ODBC SQL
Contact

Let's Connect

I'm open to roles where one person owns real surface area where software meets an industry vertical, the kind of work where directing agentic AI lets me ship like a full team.