The observation
AI changed how engineering teams work. The way teams see that work hasn’t caught up.
Sessions are happening across Claude, Codex, Cursor, and other tools. Dollars are being spent without attribution. The same kinds of waste recur in every workspace — bloated memory, looping sessions, wrong-sized models, dead-end exploration — but nobody can name them yet.
Poppi is the layer that names them.
Who’s building this
One engineer, the same loop he’s spent a career closing.
Poppi is built by Shawn McKay — senior staff engineer and DevX lead at Amplitude. Earlier: two startups whose platforms grew past a billion evaluations.
Most of that work has been the same shape: instrument the thing, measure the outcome, close the loop between the decision and the result. AI-assisted coding is where that loop is widest right now.
What we believe
Four principles we’re building around.
-
The thing you measure is the thing you can change
AI coding sessions deserve the same observability discipline as any other production system. Right now they have less.
-
Reports beat dashboards
Engineers don't want another tab. They want a ranked list of what's worth changing this week, delivered between blocks of work, not during them.
-
Your prompts and your code stay yours
We do the analysis on metadata — tool calls, token counts, model choices, PR outcomes. Deeper inspection only on the repos you opt into. Never used to train, never resold.
-
Pre-launch is a posture, not a void
We share what we're learning as we build. Design partners get direct input on the roadmap, not just a free seat.
Get in touch
Join the first wave — or just say hi.
Design-partner slots are limited and shape the roadmap. If you’d rather talk first, the LinkedIn link is the fastest way to reach Shawn.