Lead Systems Engineer

Jon Allyn

I turn infrastructure clickops into peer-reviewable code. My team now writes 100% of it with AI, spec first.

Visits
unavailable
Status
no probe data
Availability
collecting
Response time
average
AWS spend
collected daily

Live figures from this site's own infrastructure — how that works.

What I do

I automate on-premises data center infrastructure for a large automotive manufacturer. At this scale a manual process is not just slow — it is an outage with a delay on it. Most of my job is converting things people click into things people review.

I got here sideways. I started scripting fixes in internal IT, moved into running large-scale compute in data centers, and now I write the code that runs it instead of running it by hand. The through-line is a low tolerance for work a machine should be doing.

100% of our code is written with AI

I lead a team that develops entirely through AI, spec first. The specification is what we argue about; generating the implementation is the cheap part. That only works because the verification underneath it is serious — tests, type checking, scanning, drift detection, and a deploy that refuses to claim success until it has confirmed the live site is serving the new commit.

If that sounds like a claim worth checking, it is. Here is how it actually works, and here is the verification running on this site.

Selected work

Problem, approach, outcome. Full write-ups on the projects page.

DNS records as code, wired to change management

Turned manual DNS clickops into a peer-reviewed pull request, with ServiceNow change records kept in step.

  • Terraform
  • Python
  • GitHub Actions
  • ServiceNow

Ownership metadata that maintains itself

Automated tagging linking workloads and components to the teams and portfolios that actually own them.

  • Python
  • Ansible
  • GitHub Actions

Six-year hardware refresh forecasting

Automated the financial forecast behind hardware refresh planning, cutting a multi-week process to days.

  • Python