Before the platform, before the fleet, before the headset — there were embers.
The record is quiet on purpose. One hundred and ten commits across half a
year: a tooling monorepo learning to run per-service pipelines
(99c7b77), a voice prototype taking shape in Python, a vault figuring out
how to hold a portfolio of work that did not exist yet. This is the part of
an adventure nobody films.
For the record, and briefly: before any of this, the U.S. Army. Then a decade of enterprise software — test platforms at USAA, refining systems at Marathon, an independent practice, modernization at Kalkomey, SRE and MLOps at Launch. By February 2026 the day job is Senior Consultant at Oteemo. The question is what the nights are for.
The answer arrives slowly. Not as a product idea — as a pattern. Every employer’s stack had the same weakness: the interesting systems were assembled from parts nobody could explain end to end. Agents, models, sandboxes, dashboards — each one half-integrated, each one someone else’s black box. The quest takes shape as a refusal: build the whole object. Orchestration services, sandboxed execution, agent catalogs, observability that keeps it honest.
No code for that yet. Just the decision to go.