Z/YOCUM
( technical program manager × product designer ) ( product designer × technical program manager )

I ship AI programs that tell you when they're wrong.

I design AI that admits what it doesn't know.

Technical program manager and product designer. Three end-to-end AI programs — governance, evaluation, and 0→1 launch — built with the eval harnesses, hard gates, and honest readouts that make them trustworthy. Salesforce · Jira · Sheets · Figma.

Product designer with a program manager's discipline. Agentic commerce, health, and home renovation — research through design systems to working prototypes, with the uncertainty made visible.

See Conduit → See The Trusted Agent → Download resume
Cadence AI-in-the-loop workflow swimlaneAI-IN-THE-LOOP · GUARDRAILS AS DESIGN
Hearth eval scorecard showing 80% adversarial safety pass rateEVAL HARNESS · SAFETY GATE FAILED AT 80%
Cadence pilot analysis showing the confidence interval crossing zeroPILOT ANALYSIS · THE CI CROSSED ZERO
Cadence program board, seven phases done7 PHASES · SHIPPED
REAL ARTIFACTS — NOT MOCKUPS
Trusted Agent intent composer
Recomp dashboard with the AI insight headline
Trusted Agent cart proposal with confidence dots
SHIPPED SCREENS — FIGMA TO PROTOTYPE
SCROLL — THE PIPELINE STARTS HERE ↓
SCROLL — THE PROCESS STARTS HERE ↓
( 01 — the problem )

Nine teams. Nine spreadsheets. Zero visibility.

That's how a multi-family solar pipeline was actually running — status re-keyed by hand, handoffs lost in email, and no one able to answer "where is this project?" So I built the program that fixed it.

An agent that shops for you is useless if you can't trust it to.

Hand a list to an AI and it swaps your brand, blows your budget, and guesses when it should ask — silently. The Trusted Agent is a study in delegating without going blind: the work visible, confidence readable, the human one tap from the wheel.

( 02 — featured program ) ( 02 — featured product )

ConduitCASE STUDY

A fragmented pipeline, unified onto one Salesforce system of record — discovery through rollout, run as a real program.

The Trusted AgentCASE STUDY

An agentic grocery app where the AI does the shopping and the human keeps the wheel — designed, systematized, and prototyped end to end.

0
EPICS, DISCOVERY → ROLLOUT
0+
STORIES, ALL TRACED TO DONE
0
AUTOMATION FLOWS LIVE
0/13
UAT CASES PASSING
0→1
CONCEPT TO WORKING PROTOTYPE
10
KIT COMPONENTS · 3 GOVERNANCE TIERS
4
SCREEN MISSION — EDGE STATES FIRST
LIVE
INTERACTIVE REACT PROTOTYPE
E1 DISCOVERYE2 PROCESS DESIGNE3 SALESFORCE BUILDE4 DATA MIGRATIONE5 AUTOMATIONE6 REPORTINGE7 UATE8 ROLLOUT
P1 RESEARCHP2 PERSONASP3 IA & FLOWSP4 WIREFRAMESP5 DESIGN SYSTEMP6 HI-FIP7 PROTOTYPEP8 HANDOFF
Read the case study → ▶ Watch the walkthrough Read the case study → ▶ Watch the film Try the prototype ↗
( 03 — the judgment layer ) ( 03 — the craft layer )

Anyone can ship artifacts. The job is the tradeoffs.

D-06

A crosswalk table, not nested logic

Remapping legacy values is a business decision — so it lives in a readable table anyone can challenge, and unmapped values fail loud instead of passing silent.

D-09

Validation stays on during the load

Disabling the guardrails to force known-bad data in would recreate the exact problem the program exists to fix. Flagged rows get remediated, not smuggled.

D-03

Before-save vs. after-save, split by job

Two automations fire on one event — separated because updating a record's own field and creating a task are different operations Salesforce optimizes differently.

Pretty is easy. Clear is the work.

PRINCIPLE 01

Insight over data dump

A chart without a takeaway is homework. The AI companion explains the trend and the next step — the product's whole reason to exist.

PRINCIPLE 02

System before screens

Tokens and components come before pixels. Consistency isn't polish — it's how an interface becomes learnable.

PRINCIPLE 03

Prototype the habit, not the demo

The flows people repeat daily get designed and tested first — retention lives in the boring screens, not the splash page.

( 04 — the roadmap )

Three programs. Three proofs.

● COMPLETE · OCT 2024 – MAR 2025

Conduit

The builder — nine teams onto one system of record. Salesforce build, migration, automation, UAT, rollout.

Read the case study → ● COMPLETE · APR – DEC 2025

Cadence

The leader — the gate said GO and I recommended against scaling on it. AI governance, evaluation, an honest readout.

Read the case study → ● COMPLETE · FEB – JUL 2026

Hearth

The launcher — a 0→1 AI-native product: PRD, eval harness, staged rollout, powered experiment.

Read the case study →

Three products. Three proofs.

● LIVE · DEC 2025 – FEB 2026

The Trusted Agent

Agentic shopping with a glass box — the AI shops, you keep the wheel. Case study + working prototype.

Open → ● COMPLETE · FEB – APR 2025

Recomp

A recomposition tracker built on one insight — it looks like a tracking problem and it's really a retention problem.

Read the case study →
○ IN PROGRESS

RALLY

Social UX for a live-service co-op game — matchmaking you can read, played on a real controller.

Coming soon
( 05 — about & contact )

Let's build the next one.

I work at the seam of two crafts: program management — Salesforce, Jira, data migration, automation, rollout — and product design — research, systems, and prototypes in Figma. All of it independent, freelance work: scoped, shipped, and documented end to end. My day job in operations keeps me fluent in how real teams and real data behave — the instinct behind everything here.

Email me LinkedIn ↗ Resume (PDF)