Scaling engineering at Echo
Building the platform under a lean AI startup: AWS/Kubernetes infrastructure, authentication, integrations, Echo MCP — and a culture that treats code as a liability.
This write-up is being expanded. The structure and the argument are here; the specifics are still being filled in.
Context
I joined Echo in December 2024 as the founding platform engineer. Echo is an AI startup that keeps its engineering team deliberately small: engineers own products end to end — planning, building, shipping, operating — instead of handing work between functions.
My slice of that is the platform itself: the AWS and Kubernetes infrastructure everything runs on, platform architecture, authentication, third-party integrations, GenAI adoption across the org, and Echo MCP — the platform’s Model Context Protocol surface.
Problem
The default failure mode for a startup at this stage is to hire ahead of need and drown in coordination. The opposite bet — staying lean — only works if you remove everything that makes small teams slow: novel infrastructure nobody has operated before, speculative abstractions, and handoffs between people who each hold a fraction of the context.
For the platform that meant one engineer had to make infrastructure a non-topic: if product engineers ever think about Kubernetes, authentication flows, or how an integration gets wired, the lean model breaks.
Approach
Three things carry most of the weight.
A boring stack. Battle-tested technology, chosen for how well it is understood rather than how interesting it is. Code is a liability — every line has a carrying cost — so the default answer to “should we build this” is no, until the problem forces a yes.
Engineers as product owners. No requirements thrown over a wall. The person who talks to the problem is the person who ships the fix, which collapses the loop between noticing something and correcting it.
AI where it is leverage. Planning and boilerplate are AI-assisted; judgment stays human. The point is not writing more code faster — it is spending engineer attention on the decisions only an engineer can make.
Result
The model held up through the company’s $50M raise, with an engineering team a fraction of the size usually attached to that number.