Product Guy. 12+ years building products that end up becoming the company.
Three product lines from zero. One company-saving pivot. Billions a year
in payments running through software I own. Good choice on the door, by the way. Nobody
verified you're human, but my fit-check agent doesn't mind either
way.
TLDR
I've spent twelve years building software for people the tech industry mostly ignores:
construction crews, food distributors, sales reps who live on their trucks. What they do
isn't sexy, so tech skips them. That's the opportunity. These industries run two-thirds of
our food supply on phone calls and paper, and product work that's table stakes in consumer
tech is a superpower there.
At Pepper (Series C, backed by ICONIQ, Index,
Greylock, and Lead Edge) I was the first product hire. Six years later, the product lines I
launched make up most of the company's revenue and are named in the
Series C announcement.
Before that I ran a product line inside WeWork, rolled out to construction teams in 15
countries in 7 languages, handling hundreds of millions of dollars in construction value.
I'm looking for Director+ product roles in New York at companies serious about agentic
AI. Not the press-release kind. The shipped kind. The kind people actually use and get
value from.
On finding the money:
At the end of 2024 I told our exec team we were leaving money on the table in payments and
asked to re-staff the line I'd built and parked. Revenue nearly tripled the next year.
Earlier versions of the same move: a $24M efficiency save at WeWork, a COVID pivot that
went from zero to $200K a week in three weeks.
On org design:
I split our engineering organization into the two-team structure the company scaled on.
I've launched three product lines, hired my own successor for two of them, and moved on to
the next bet. Today I run a leadership pod (product, engineering, design) that operates
independently because it earned the org's trust.
On pricing:
Our payments product used to be bundled free into contracts. Nobody used it. I convinced
our own sales team the product was worth charging for, and paying distributors activated
where free ones never had. Subscription revenue went from zero to a real share of the line
in 19 months. Pricing is a product feature.
On AI:
I was building ML features before the hype, and I'll tell you honestly why my 2024
demand-prediction feature didn't matter (the post-mortem is below). Now I build agents
output-first: generate the deliverable from real data, put it in front of a customer, and
only build the software after they tell you it's worth paying for. My collections agent
was a PDF before it was a product. That's the point.
The story
2014 to 2017, Fieldlens. Talked my way into product with no PM experience by
convincing the head of product I could do the job, then earned the core product. Helped
grow ARR from $50K to $1M+ toward an acquisition by WeWork. Won a 2015 Gold Stevie for
Best New Technology against apps from Accenture, TriNet, and Capital One.
2017 to 2020, WeWork. Arrived via the acquisition. The Fieldlens executives
scattered into the larger org and the entire product line landed on me alone, inside the
department building technology for real estate and development. Rolled it out to
construction teams in 15 countries in 7 languages, built the metrics WeWork used to
measure build speed and quality on 700+ openings, roughly $24M in measured construction
efficiencies in 2018, and kept every external client through the acquisition with zero
churn. Then WeWork happened, for reasons famously unrelated to my corner of it, and I
went looking for the next industry tech forgot.
2020, the pivot. I joined Pepper in March 2020 as the first PM. Within weeks COVID
zeroed out our restaurant customer base. We built Pepper Pantry, direct-to-consumer sales
for food distributors' stranded inventory, and hit $200K weekly revenue in three weeks. It
kept the company alive. That summer I read about Shopify arming the rebels against Amazon,
sent the article to my CEO, and that became our white-label strategy: don't build a
marketplace that makes distributors compete on price, give each one their own storefront.
First contract, August 2020.
2020 to 2024, the foundation. Sole PM for four years. Built the Ordering
platform: 115 distributor launches, 20K active users. Launched payments as a second
business. Split engineering into two teams. Grew the team from 3 to 12 engineers.
2024, the portfolio year. Handed Ordering to our second PM. Cannibalized the
payments team on purpose to seed Sales Hub, tools for distributor sales reps. Took it from
zero to real revenue in months, hired my own replacement to run it, then took the next bet
(sales managers). Then went back for payments.
2025 to now, Finance Hub. Re-staffed payments, expanded it into Finance Hub,
launched it in Canada and Australia, added the subscription layer, and nearly tripled
revenue in year one. Processing volume more than doubled in the last twelve months. And I
ship AI agents (next section).
AI: receipts, not vibes
My thesis: agents are output-first products. An agent's value is doing work on a
human's behalf, either suggesting actions a human approves or deleting a workflow
entirely. So don't build the system first. Generate the output from data as real as you
can get, show it to a customer, and let them tell you whether it's correct and whether
it's valuable. AI makes this loop very fast. Sell the output, then build the machine.
Longer version here.
On item recommendations (early Pepper, still in production):
The first ML feature I shipped, years before the hype. Ordering suggests items a
restaurant should buy based on what similar restaurants purchase: you buy ketchup and buns
but not mustard, we suggest mustard. It got good sales traction and it's still live. My
honest grade: relatively good, not amazing. The v1 model ignored geography, restaurant
type, and substitutes, and we refined it later. First lesson in the gap between "the model
works" and "the model makes money."
On Expected Orders (2024), the one that didn't matter:
Pre-hype ML feature predicting which restaurants would order today, built for sales reps
who ran on notes and memory instead of documented behavior. Shipped Q3 2024. Clients found
it interesting. It didn't change behavior. Post-mortem: the problem wasn't painful enough,
and the 40-order data threshold meant our newest distributors, the most engaged ones,
couldn't use it during ramp-up. I chose not to rebuild it. Knowing when an AI feature
isn't worth it is the skill the hype cycle forgot.
On the collections agent (2026):
AR collections is a grind distributors do badly. I generated the agent's work product,
collection plans built from real (anonymized) receivables data, as a PDF and pitched it to
distributors before any software existed. Their feedback shaped what we're now building as
a web app. Demand-validated before a line of app code. (Revenue numbers will show up here
when they're signed, not verbal.)
On the prospecting agent (in flight):
My direct report is building an agent that prospects for distributors: finds nearby
restaurants on the open web, filters out existing customers, ranks leads by fit (a pizza
distributor gets pizzerias, not Mexican restaurants), pulls prospect menus to generate a
quote to walk in with, then runs the pipeline like a CRM that follows up with the rep and
suggests the next move. My job is coaching him through the playbook. At this level the
playbook transferring is the work.
On the autonomous dev pipeline:
I built my team an AI pipeline that takes a ticket from our queue, classifies it,
implements it, writes tests, reviews itself, opens a labeled PR, and updates Jira. I'm a
product director, not an engineer. This site, its agent, and that pipeline were all built
the same way.
Here's why I care: every product and engineering team needs to stay ahead of the
frontier of what AI can realistically build. Today that frontier covers smaller,
well-scoped work. It will keep expanding into larger and more complex projects. For years
PMs have had to sacrifice the incremental improvements (the papercuts, the polish, the
small asks that keep customers happy and keep a product excellent at its job) to fund the
big innovative bets. AI deletes that tradeoff. The incremental layer increasingly takes
care of itself, so teams get to pursue bigger bets without letting the product rot. The
best teams of tomorrow are already watching this frontier today.
Hiring? Paste a job description and I'll tell you honestly if we're a fit.
You get a structured verdict: where I'm strong for the role with sourced evidence, where
I'd genuinely stretch, and the questions to ask me in an interview to pressure-test those
stretches. No job description handy? You can also just chat with the agent that
represents me: the work, the numbers, the gaps, the corgi.
The agent is instructed to admit gaps. If it says I'm a fit, it means it.
Personal
Born and raised in Guaynabo, Puerto Rico, until 18. Boston for college, New York ever
since. Brooklyn now. Bilingual, English and Spanish, both native. USA Cycling Category 1
road racer (one step below professional). Licensed private pilot. Tufts cognitive science,
which turned out to be decent training for working with LLMs. A corgi named Potato. A
daughter.