Washington, DC

Daniel Fenton

Leader. Builder. Teacher.

Built — JLL-GPT, 25K daily · ArcGIS Hub, 27K sites · Blackbird, global · Koop, open source

I build JLL's enterprise AI platform, used daily by about twenty-five thousand white-collar employees to do their job. I lead a thirty-person team supporting JLL's hundred-and-ten thousand employees. I build directly with Claude Code and Codex, and I teach the people around me to do the same. On nights and weekends I ship Claude agents for my own life: a kitchen, a garden, a pen plotter.

The throughline is two industries, commercial real estate and civic tech, with platforms as the consistent unit of work. Before JLL-GPT I launched ArcGIS Hub and Open Data at Esri, now powering more than twenty-seven thousand sites. Before that I built Blackbird at JLL, one of the first analytics platforms in commercial real estate, still in use a decade later.

The site is mostly the rest of that. Work that doesn't fit on a resume.

Right now

Day job

JLL AI Platform

2023 — present

I was an author of JLL's 2030 strategy. Built JLL-GPT and JLL Falcon, the platform underneath. JLL has calculated nearly $200 million in productivity gains from JLL-GPT across the org; about twenty-five thousand white-collar employees use it daily to do their job. The work has made JLL the leader in AI in commercial real estate and one of the AI leaders among Fortune 200 companies.

The architecture: tiered agents (Tier 0 high-code embedded with line-of-business engineering for facilities, leasing, capital markets, property management; Tier 1 and 2 for no-code citizen development), an MCP gateway for tools, an LLM gateway for model routing and observability, governance on Unity Catalog and Databricks.

Solo-built alongside the leadership work: Pulse (an agent scaling rapidly across JLL) and an MCP server extending our AI applications with JLL data.

Earlier work in this stint: LeaseLens for commercial lease ingestion, Custom Assistants versioning, the JLL-GPT name itself.

JLL Falcon announcement →

Side project

Chef Claude

2026 —

A personal chef for the Fenton household. A long-running Claude agent picks the week's recipes and the grocery list. It takes feedback and learns what we actually like. We cook from it almost every night. Native iOS on the phone, FastAPI on AWS underneath.

github.com/dmfenton/chef →

Chef Claude, weekly meal planning screen on iOS

Things still running

Software I made or led that's still in production. The work I'm proudest of, mostly because it lasted past my interest in it.

ArcGIS Hub & Open Data

2014 — 2019

Open Data DC, a representative ArcGIS Hub site

Launched ArcGIS Hub and the Open Data product line at Esri. It powers around twenty-five thousand government sites today: counties, cities, agencies, all of them making public data browsable and downloadable. The District uses one (opendata.dc.gov, pictured); so do most of the others. I designed the search, picked Elasticsearch, and ran the team that scaled global ingest from eighteen thousand downloads a month to over a million.

opendata.dc.gov →
Blackbird

2013 — present

Blackbird, Manhattan filtered by walk-time to the subway

Created it as a product manager at JLL. One of the first analytics platforms in commercial real estate, before "data platform" was a default category. Now in use globally at JLL. Came back later as Senior Director of Product Engineering to set its next chapter; the simplified 3D visualization we shipped delivered an immediate fifty percent increase in usage.

"The worst thing for a product team is developing something that people don't use."
me, in a Pendo case study about Blackbird

pendo.io case study →
Koop

2015 — present

A developer-friendly API interoperability library: an open-source toolkit that turns any data source into a geospatial API. Shipped at Esri; still in use today.

koopjs.github.io →

How I work

Builder PM. I lead, and I build. Claude Code and Codex are the daily drivers.

Platform before feature: a central orchestrator, an MCP gateway, an LLM gateway, a tiered development model, then a governance fabric. Features come after the chassis works. Scenarios first. Every agent track is anchored to a concrete business scenario before code is committed.

A bias toward physical artifacts. The most interesting agent loops close on something real: a drawing pulled by a plotter, a valve opening in the garden. Side projects pick a physical endpoint on purpose.

Communication: clinical, concise, substance over fluff. Slides and memos are precise strings, not paragraphs.

Other things I've made

The showpiece below. Smaller weekend experiments live in the lab; roughly ninety more sit in github.com/dmfenton.

Code Monet

2026

Untitled study, generated by Code Monet

Open-source autonomous AI art system. Claude agents drive pen plotters, see screenshots of their own output, and iterate on the work. A closed-loop agent: plan, actuate, perceive, replan. Demonstrates that an agentic loop can close on physical artifacts, not just text.

monet.dmfenton.net →source →

And the hardware bench: a GPS cat tracker on Helium, a wildlife camera, dev-board work on the AMB82-Mini NPU, HDR pipelines on Picamera2, occasional servo repair on a small robot. The building habit doesn't stop at software. See more in the lab.

Where I've worked

2019 — present

JLL · Head of Product, AI Platform

Built JLL-GPT and the platform underneath.
2014 — 2019

Esri · Lead Software Engineer

Launched ArcGIS Open Data; led the platform that became ArcGIS Hub.
2011 — 2014

JLL · Product Manager

Built Blackbird, geospatial intelligence for commercial real estate.

Education

University of Maryland, Computer Science: geospatial computing, information retrieval, human-computer interaction.
Georgia Tech, MS, Applied Economics.
College of Charleston, BA, Economics and Philosophy.

Patent

US #14516893, Computing system for visualizing integrated real estate data.

People I've grown

I've hired and grown a series of product managers across JLL and Esri. A handful of past directs now lead teams of their own.