top of page
Frida_edited.jpg

RENEE ROMERO

AI Systems Designer​

I design where agents stop and ask a human.

Framework

Human Checkpoint Placement

A framework for finding out whether a checkpoint in an agentic system actually works. The 15 Agent Checkpoint Heuristics, starting with my own system.

hcp_card 1.png

FRAMEWORK

Where agents stop and ask a human

Scope is a boundary you draw before an agent runs. Verification is a check you run after it finishes. Between them is the part almost nobody builds.

Projects

Projects
Chat.png

Agentic Support System

A customer said his AI couldn't help him because it didn't know my system. He was right. So I built a 13-agent support system on Cloudflare Workers, grounded in the real workflow code, answering paying customers on demand.​

Play case study
View case study
Agentic_Thumbnail.png

Agentic Job Search Automation

A 9-workflow agentic system that cut drafting from about 90 minutes to 15 per application, monitors my target company list daily, and auto-matches 500+ contacts to roles. 13 paying customers, $2,800+ in revenue, and a Cultivated Culture podcast invite.

Play case study
View case study
Thumbnail laptop Scribbler logo.png

AI for Officer Efficiency

Officers left the letter editor to find legal citations, and paste-backs broke formatting. A stakeholder's dormant cost report put the gap at $372K to $1.1M a year. I specified Scribbler, an in-editor retrieval layer, and the design lead validated it as the next build.

Play case study
View case study
Teardowns

Teardowns

I take AI products apart to see where they let a human step in before something goes wrong. I didn't work at any of these companies. I used what any customer can see, then scored what I could on the 15 Agent Checkpoint Heuristics.

Linear Logo.png

LINEAR

Agent write flow

The agent tells you exactly what it changed, within seconds. It just doesn't ask first. Linear already shows everything the agent picked. It just shows it after, not before.

Supabase logo.png

SUPABASE

MCP confirmation flow

It asks you to confirm a change based on how risky it looks, but never checks whether the change is correct. A broken security rule looks like any other change and goes live. Written before the 15 Agent Checkpoint Heuristics existed. Not yet scored.

© 2026 by Renee Romero

bottom of page