Matt Kundo

Marketing operations · GTM engineering · Austin, TX

Marketer who builds the internal tools.

Ten years across paid media, SEO, analytics and CRM operations, engineering the measurement and automation since 2024 rather than doing it by hand, and since May 2026 a production Python platform on the Claude Agent SDK. Energy is the vertical I know best: ERCOT, TDU pricing and enrollment funnels. I am winding down my consultancy for an in-house role.

571

pull requests merged in 105 days, one person and the platform

393

agent-callable tools I built and operate

101%

measured ROI on a $76,279 channel mix

7,141

tests passing, which is how the speed stays honest

Download resume (PDF) Email me

writing linkedin.com/in/mattkundo github.com/mattkundo texascommercialplans.com

The platform

A production autonomous agent system in Python that I built to work alongside me across my client book. Postgres with pgvector semantic memory, deterministic compliance gates, database-backed kill switches, and a completion gate that requires evidence before a task can be called done.

It does not run the operation. It plays critical support, and it extends what I can deliver. Client email is draft-only except the weekly performance report, which auto-sends only when I arm it. Monthly reports need a manual confirm before release. Client ad budget, bid and copy changes are blocked outright. One account has a narrow, kill-switched autopilot that can only pause spend, never raise it. Stating the limits is what makes the rest credible.

scale237,268 lines of Python in the repo, tests included, 103 scheduled jobs, 29 agent roles

measuredas of 2026-09-02, derived from the repository rather than typed by hand

Results

101% ROI on a $76,279 channel mix for a retail electricity provider, October 2025. $153,610 revenue and 594 enrollments, closed customers rather than leads, at a blended cost of $128.45 each, measured through an attribution model I built.
$253 to $38 cost per conversion after launching paid search from zero for an at-home health testing e-commerce brand, monthly conversions from single digits to 39.
Two Google Ads conversion actions at zero for their entire lifetime on that same retail electricity account, so bidding was optimizing toward a signal that didn't exist. I built the uploader that fixed it: 194 of 197 accepted, $48,570 in conversion value.
42.6% conversion rate on the stronger of two rebuilt paid landing page variants, against 23.3% on the other.

Clients anonymized by agreement. Every figure traces to a named system of record.

The arc

2015 to 2018Paid media and SEO, agency and in-house

2018 to nowFounded MKDM. Over $75K a month in ad spend across up to 15 accounts at the book's peak, carried continuously since 2018

2026 to nowBuilt the agent platform. Shipped texascommercialplans.com solo: scraper, database, site, content engine

nextIn-house, where I can build one system properly instead of five in parallel