Who I am
On call for everything, by choice.
I'm Ethan Jensen, founder of Autonomous Interactions. The fastest way to learn operations is to be responsible for all of it — so I built a private cloud most small companies would recognize, then built the AI layer that runs it.
Production discipline, personal scale
I run a multi-node virtualization cluster, redundant ZFS storage, directory services, a self-hosted ERP, GitLab, a SIEM, and full observability — versioned, monitored, ticketed. Then the interesting part: an autonomous operations layer that watches the alerts, files and deduplicates tickets, drafts maintenance plans, and — behind explicit approval gates — executes repairs and validates the result.
My belief, demonstrated rather than argued: a small operation with good harnesses can run infrastructure that normally takes a team, without renting a single seat of SaaS. When something breaks at 2 a.m., the postmortem is mine to write — and that accountability is the difference between knowing a tool's marketing page and knowing its failure modes.
Operator profile
Comfortable from SAS PHY error counters to LLM orchestration — most useful in the layer between, where they have to trust each other.
Trajectory
How the lab became a platform.
Build the substrate
Five-node virtualization cluster, dual ZFS NAS, domain services, GPU hosts, a zero-trust tunnel for anything public. Boring on purpose — the substrate's job is to be forgettable.
Earn the operational scars
Disk-fills masquerading as GPU bugs, a gateway brought down by mass container churn, a migration that silently destroyed data on a mis-wired disk shelf. Each one became a written rule. The rules became the method.
Replace the rented stack
ERP and helpdesk, GitLab, secrets vault, SIEM, workflow automation — self-hosted, integrated, license-free. A collection of apps became one system with one source of truth.
Close the loop
Alerts file tickets. Tickets become plans. Plans book maintenance windows. An agent proposes repairs with blast-radius estimates, shadow-graded until accuracy earns the execute path. The first operator-approved autonomous repair has shipped.
Generalize it
The harnesses, the world-model, the planner — none of it is lab-specific. It's a blueprint for how a small team runs serious infrastructure with AI doing the toil and humans holding the gates. That's the work I want to do for others.
Operating values
What you get when you work with me.
Honest reporting
If the test failed, you hear "it failed" with the output — not a softened summary. Decisions made on flattering data are expensive.
Receipts, always
Diagnoses come with the exact output that confirmed them. Claims about systems are verifiable, or they're not made.
Calm under incident
Triage order, not panic: check the boring thing first, snapshot before mutating, never compound a failure with an unverified fix.