Agent orchestration
01I run a standing team of specialist agents that spec, build, review, and verify software. One instruction fans them out in parallel and they hand work down a real pipeline.
proof: Field Log AI-native product leader · running a live agent org
I build the systems
that make large organizations
ship like small ones.
Product leadership plus a personal working AI practice. AI isn't just what I do AT work. I orchestrate teams of agents and ship real software with them that is changing my life and the lives of real people around me.
Brandon Smith
Director, Lead Product Manager of the Enterprise Operating Model
Evernorth Health Services (Cigna)
$ eugene dispatch "portfolio-v2 --parallel" prism design running forge build queued scout verify queued herald comms queued ✓ 16 agents online · deepgram nova-3 · claude brain
AI capability // receipts, not adjectives
The skills, with receipts.
Named plainly, the way the industry names them. Each one has a build or a result behind it, not a bullet.
Multi-agent systems
02Eugene dispatches detached workers over a stdio-MCP bus. Each agent carries a tuned role, its own toolset, and a permission boundary.
proof: Eugene agent orgPrompt & context engineering
03Persona-based toolkits, plus a self-knowledge layer that renders an agent's own code into its prompt and drift-checks it on every commit.
proof: self-knowledge layerEvals vs. pre-AI baselines
04Every pilot starts from a measured pre-AI baseline and ships with evals. I report lift against the baseline, not demo-day vibes.
proof: pilot methodAI-assisted prototyping
05Eight production apps in daily use, built end to end with an AI toolchain. Discovery to deploy on real stacks, shipped to real users.
proof: the builds belowVerification & evidence
06Evidence over assertion. Render-and-probe checks, live smoke tests, and defect ledgers gate anything before it ships.
proof: verify tooling The spine // product management
The discipline that makes the AI actually land.
AI without adoption is a demo. Product management is how a working model becomes something people rely on. Four questions decide it: is it valuable, is it viable, is it usable, and can we actually build it.
ValuableDoes it solve a problem someone actually has? Every build starts at a real person's highest-friction work.real need
ViabilityDoes it hold up beyond the demo? Pilots run on production AI and earn their keep against a measured baseline.beats baseline
UsabilityCan someone use it without a manual? I ship to real users and watch what they do, not what they say.real users
FeasibilityCan we build and run it for real? I prototype with the same AI tools my team uses, so the answer is working software.working software
01 · hud 
02 · agents 
03 · mission control 
04 · tool feed 
build // 02
Hearthbook
A 5e player companion: character, prep, rules, recaps.
Session table view
Character sheet
SRD rules lookup
build // 03
Karaoke
A live sign-up station and big-screen queue for a real night.
Queue position, live
Live song search
Big-screen now playing
build // 04
Ember
A private app for two people, where the privacy rules are the product.
Shared streak and state
Both answers, unlocked
Signals and nudges
The rest of the shelf
More shipped, same discipline.
Mission Controlprivate
A personal ops dashboard: calendar, tasks, and status in one view. data reconciliation · scheduled jobs Packing List
A smart, reusable packing app that learns a list per trip type. discovery · offline PWA Field Loglive
The public record of the agent team actually working. AI orchestration · evidence OpsPilotsample
A purpose-built job tracker for a small HVAC business. requirements from scratch · reporting cutprivate
A nutrition and weight app for two users, food and barcode lookup. 3rd-party API · trend modeling