Co-founded Prompteus with a few people tired of watching AI projects fail at the enterprise level. We built an LLM gateway for regulated industries: routing, caching, observability, and the governance and audit pieces that enterprises need but nobody wants to build.
Two thousand developers signed up in the first week. Six figures in ARR by end of 2025. Good signal.
The other good signal came during that same first week. An investor we respect reached out, dug into what we were doing, and told us it would get commoditized. LLM gateways are the easy part. Routing, retries, caching, a bit of observability. Useful, not defensible. He was right.
So we spent the rest of 2025 figuring out where the defensible part actually lives. It sits around the model, not inside it. The industry now calls this the harness: the part that captures the tribal knowledge operators hold in their heads, enforces deterministic rules at the system boundary, and still lets models correct for the hundred small things a human would fix without thinking.
Most AI projects die between “cool demo” and “production.” Not because the AI doesn’t work, but because nobody built the boring infrastructure parts. Cost control, audit trails, data governance, proper caching. We built that at Prompteus. Then we learned the harder, more defensible version sits one layer up.
Prompteus is the gateway. RakerOne is what we built on top of it — the operational layer where AI actually runs the work, inside rules that hold. Same crew, same thesis, one altitude higher.