Fandangling…

AI product leader. Agent expert.

I lead product orgs, own the P&L, and run a production AI organization that ships while I sleep: 30+ agents in daily production, 100+ built over time.

BYU MBA class of 2027. Open to AI product and applied-AI leadership conversations now.

See the work ↓ Teardown → Resume ↓ GitHub ↗ Book a meeting →
$20M+
business scaled, from $2M
5→35
engineering org grown
38
agents on the public roster
587M
tokens processed yesterday
13,256
users across shipped products (Aug 2026)
SNAPSHOT · 2026-08-28 · every build below is a live link or a clone-able repo
JD Davenport
The flagship build

The Nerve Center: a production AI organization

One CEO agent fans out to specialists across my life and work domains: 30+ agents in daily production, 100+ built over time. Roughly 200 durable workflows, a live memory graph, every layer designed and built by me, including the design system on this page.

nerve center · memory graph · june 2026
Nerve Center memory graph, an amber galaxy of the agent org's memory nodes
System telemetry snapshot · 2026-08-27

587M tokens and 4,794 turns yesterday. Work volume through the system, list-price equivalent.

2026-07-2930 days2026-08-27
140
days in daily production, since 2026-04-10
3 →
incidents documented, with root causes and fixes
16,086
tests in the estate, one per guardrail that matters
40 ↗
public repos on jddavenportOpen
Token volume is a list-price repricing of local transcripts: what the work would cost at published rates. It is not a bill and not plan consumption, which are different measurements.

It runs in production, which means it breaks in production. I run it anyway, behind machine-enforced guardrails and a send gate that stops it. If you want the architecture, the failures, and the fixes: read the production teardown → The system also publishes its own census every day: see the live capabilities page ↗

Shipped & clickable

Things I've built. Clone them, click them.

No slideware. Every item here is a public repo you can clone or a URL you can open. The first row is one arc: the story, the system it came from, and the same controls packaged for an enterprise.

The artifact

Capability paired with the risk it carries, the control that bounds it, and the incident where the control failed anyway. Read the case study, then run the code it describes.

Case study · start here
agent-safety-case-study public

Every capability this organization has, the concrete risk it carries, the control that bounds it, the tradeoff I accepted, and one incident where the control failed and what changed as a result.

github ↗ read the teardown →
Open source · MIT · Python
orchestra-agents public repo

The production core as installable code: fleet fan-out under bounded concurrency, a durable message bus that dead-letters loudly, circuit breakers, best-of-N with an adversarial judge, and permission tiers from read to irreversible.

github ↗ git clone & run
Open source · MIT · Python
claude-deploy-kit public repo

The same controls shaped for an enterprise: one policy chokepoint that defaults to deny, a destructive-command hook that holds even under skip-permissions, a fail-closed eval gate, and a redacting audit log.

github ↗ git clone & run

The system behind it

Open source · MIT · MCP server
mcp-judge public repo

A calibrated multi-persona LLM judge, exposed as an MCP server and shipped with the eval harness that proves the calibration. It names the four ways judges fail and tests for each one.

github ↗ git clone & run
Open source · MIT
claude-bug-squash public repo

Point it at a repo and it triages, fixes in an isolated worktree, and proves red to green. Every fix clears blast-radius caps, a never-touch deny-list, and a three-seat adversarial review. Out of the box it can never merge.

github ↗ git clone & run
Production system
The Nerve Center in production

The private system everything above came out of. It runs my actual life and work, so it stays closed. The receipts are open: a full teardown, and a capabilities page the system regenerates from live state every day.

read the teardown → live capabilities ↗
Open source · MIT · Claude Code skills
recruit-copilot public repo

A job search treated as a verification problem: one experience bank built from the resumes you already have, roles scored against goals you set, a resume tailored per posting and proven machine-readable. It does not apply for you. That is the point, not a limitation.

github ↗ git clone & run

Also shipped

deep-research-agent Iterative research with a knowledge-gap loop, nine parallel sources, and per-claim refute-or-survive verification. github ↗
context-kit Four personal-context templates and five Claude Code skills, so an agent starts with real context instead of guessing. One-command install. github ↗write-up ↗
BYU AI Foundry The program I founded at BYU Marriott: student builders shipping production AI for real clients. Its jobs board is scored by resume-grader, a documented LLM judge. open live ↗resume-grader ↗
clawdling · fleetwright · harnessview The open bring-your-own-key engine behind the private org, a self-hosted fleet cockpit, and a harness inspector. clawdling ↗fleetwright ↗harnessview ↗
docs.agenttree.army Engineering write-ups and the Agent Dispatch newsletter: architecture and orchestration notes from running the org. docs ↗newsletter ↗
Track record

Product leadership at enterprise scale

8+ years of product leadership: Principal PM at a Fortune 500, Deloitte, a funded startup as co-founder and CPO, and production AI at Siemens today.

Full history on LinkedIn →
2026 → now
AI Builder · Siemens
Shipping production AI features inside a Fortune 500 enterprise software organization.
2025 → now
Founder · BYU AI Foundry
Founded and created BYU Marriott's AI Foundry: student builders shipping production AI for real clients.
2025 → 2027
MBA · BYU Marriott School
Strategy and product management. Building the MBA Office's internal software suite.
2022 → 2025
Principal PM · National Grid
Grew a fixed-price digital services business from $2M to $20M+ annualized, with an AI pricing engine as the key call. The P&L had many hands. The pricing decision was mine.
2022
Co-Founder & CPO · Dominium Finance
Raised $1.2M pre-seed; $600K revenue in the first 90 days before the 2022 Web3 downturn.
2020 → 2022
Product Consultant → Senior · Deloitte
GM, Pfizer, and Walmart engagements. Promoted to senior in year one.
2018 → 2020
PM / Full-Stack Developer · Brio Energy
Built the CRM and marketing site that drove $3.9M in new annual revenue.
2017 → 2018
Full-Stack Developer · Surge Software
First dev job. Shipped 12 production APIs in 3.5 months.
2015 → 2020
BYU · B.S. Management Information Systems
Coding, databases, systems design.
2013 → 2015
Missionary · Oaxaca, Mexico
Full-time service leader. Fluent Spanish.
Get in touch

Reach out directly

Open to AI product and applied-AI leadership roles, and to conversations about agents in production. No form.

Book a meeting → Connect on LinkedIn → me@jddavenport.com GitHub