About Eric Caskey

Senior Software Engineer at Amazon. An engineering team of one, working in the open.

Who runs this site

Eric is a Senior Software Engineer at Amazon. This site and its projects are personal work, separate from that role. It is not a company and sells nothing.

The team is one engineer and a fleet of AI coding agents, run the way I run a platform at work. Every feature starts as a spec in a versioned repository. A supervisor model plans and reviews, cheaper models type, a work queue decides what is next, and the same checks the pipeline runs have to pass locally before anything merges. Nothing publishes itself: the merge that deploys stays mine. The lesson series on this site teaches that setup one file at a time, and the numbers posts count what it produces and what it costs.

So far that is 12 case studies, 72 essays, and 9 lessons in the series, plus the working software in the Lab: a factor-scored stock board, 3D market visualizations, a marathon coach grounded in real training data, and a finance reviewer where the model narrates numbers it never derives. Off this site: Ballast, an open-source RAG system engineered to refuse when it should; SpecSelf, a spec-driven operating system for a life whose agents read but never write; and a C++ options engine tuned from 15 to 215 million prices a second.

Where the method comes from

I am a Senior Software Engineer at Amazon, on a platform team whose job is to be a force multiplier for service teams. I joined in June 2022 and built a monitoring lifecycle platform that brought 3 million monitors under standardized management across 2,750+ application stages, while hiring and mentoring eight new-to-industry engineers in New York. Since August 2024 I have owned that platform and architected the workflow orchestration control plane whose guardrails validate every change before it runs, for development teams in multiple global regions. Those guardrails have run 500,000 cumulative safety checks so far.

The part I care most about is the safety layer. A pre-execution engine runs concurrent checks before every automated operation: change control, regional isolation, monitoring readiness, pipeline health. A just-in-time composite monitor aggregates the alarm sources an execution depends on into one ephemeral, execution-scoped check, and an environment-isolation guardrail cannot be opted out of. An incorrect yes from a system like this is a production incident, so the system is built to earn every yes.

I also drove the platform's spec-driven AI strategy: a specification-as-code system as the single source of truth, and a hub-and-spoke setup where specialist coding agents inherit a curated slice of context instead of the whole corpus. The bet is that context architecture beats documentation dumps. So far it has held.

Before Amazon

Nine years at Prudential Financial, from Tier 3 help desk through Senior Remote Access SRE to a final year as a Cloud Engineer in Site Reliability Engineering, keeping enterprise VPN and MFA services quiet. That meant a VPN serving 60,000 corporate users through the early months of COVID, an MFA self-service portal used more than 18,000 times in six languages, over 300,000 activation codes from a QR enrollment tool I built as a side project in 2013 and ran for eight years, and over 200,000 administrative actions automated. The last year was hands-on AWS and Azure ownership at Fortune 100 scale, and it is the bridge to Amazon. Most of that work was removing failure modes rather than papering over them.

What I believe

  1. Guardrails, not booleans. Every guardrail I build has to say not just what failed, but why, what would make it pass, and who owns remediation.
  2. Structure before code. The specs, folder structure, and decision logs that precede implementation decide whether the system holds up; the AI follows the structure, not the other way around.
  3. Scale is the evidence. A title tells you where someone sits, not what they own; the systems, and the numbers behind them, carry the argument.
  4. Earned lessons only. I write what I have learned by getting it wrong first, in production.

Mission

Build the systems other engineers depend on. Write honestly about how it goes.

The career narrative, start to finish, is on ericcaskey.com/about.