Platform engineering · AI reliability
Eric Caskey
I make AI systems safe enough to depend on, at org scale: 3 million monitors, 2,750+ application stages.
Fifteen years making production infrastructure safe to depend on. At Amazon I architect a multi-region workflow orchestration platform with validation guardrails that run before anything executes. Now I bring that same rigor to AI: spec-driven systems, curated context, and validation that catches a wrong “yes” before it ships. I write about how this work actually gets done at Caskey Engineering.
Work
Featured work
- 500,000 automated actions a day
- 3 million monitors
Architected a workflow orchestration platform built safe by default: validation guardrails run before every automated change, and a spec-as-code system gives AI coding agents curated context instead of raw access. It runs 500,000 automated actions a day across Amazon's fleet. I also own the monitoring platform underneath it, which keeps 3 million monitors standardized across 2,750+ application stages.
Case study on Caskey Engineering →- 200,000+ administrative actions
- 60,000 corporate users
Built the MFA self-service portal used over 18,000 times in six languages, automated 200,000+ administrative actions, and kept the VPN running for 60,000 corporate users through the early months of COVID.
Case study on Caskey Engineering →Writing
Selected writing
Ballast: An LLM App Whose Best Feature Is Saying 'I Don't Know'
June 27, 2026 · Caskey Engineering
A RAG system whose most important feature is refusing to answer. How trust got built into the architecture instead of the prompt.
Designing Safety Guardrails for Distributed Workflow Orchestration
April 10, 2026 · Caskey Engineering
What I've learned building validation engines for infrastructure where an incorrect “yes” is a production incident.
Get in touch
I read everything. The contact form is the fastest route.
Contact →