Production AI engineering for San Francisco AI-native scale-ups: evals, agents, cost control, internal tooling. Senior remote delivery from the EU, per project.
San Francisco is the one market that does not need to be told what AI is. Between SoMa, the Mission and the wider Bay Area sit the model labs, the accelerators, and a generation of scale-ups whose product roadmaps assume large language models the way an earlier generation assumed the cloud. Engineers here have already read the papers, already built the prototype, and already discovered the gap that nobody talks about at demo day: the distance between a notebook that works and a system that survives real users, real cost and real edge cases.
That gap is where the actual work lives, and it is unglamorous. Evaluation harnesses so a prompt change does not silently regress. Observability so you can see what the agent actually did. Cost and latency control (model routing, caching, batching) because inference bills scale with success. Guardrails, fallbacks and human escalation for the ten percent of inputs the happy path never anticipated. And behind all of it, the internal tooling that every fast-growing company needs and nobody has the headcount to build, because the engineers you did hire are booked on the roadmap for the next two quarters.
EIGHTY8 is a Netherlands-based, remote-first AI automation and consultancy studio, and we are the team that does that work. What the Bay Area calls a forward deployed engineer (FDE), delivered remotely. We take prototypes to production, build the internal automations and agents that free your own engineers, and hold to a standard your reviewers will recognise. We are honest about the time-zone cost: the Bay Area is our late afternoon, which leaves roughly a two-hour live window against your morning, so we run async-first with a written handover waiting when you start your day. In exchange you get senior European delivery capacity. And when your EU expansion arrives, GDPR-aware engineering is not something we have to go and learn. Every engagement is quoted per project.
For capacity and for a perspective you cannot hire locally at speed. Senior engineers here are scarce and slow to recruit, and the work we take on (production-hardening, internal tooling, the automation nobody's roadmap has room for) is exactly what gets deferred. We also build with European data-protection constraints as a default, which becomes an asset the moment your first EU enterprise deal lands.
Usually four things: an evaluation set so changes can be measured instead of vibed, observability so failures are visible rather than reported by a customer, cost and latency control through model routing, caching and batching, and guardrails with human escalation for inputs the happy path never saw. None of it is glamorous and all of it is the difference between a demo and a system.
Honestly, and by design rather than by heroics. Your morning is our late afternoon, which gives roughly a two-hour live window for calls and reviews; everything else runs async. You get a written handover at the end of our day, so progress and open questions are waiting when you start yours. It suits teams that work from written specs and pull requests, and it is a poor fit for teams that need someone in the room at 4pm Pacific.
We scope and quote per project, based on complexity, after a free technical conversation. No seat-based pricing and no published rate card, because hardening an existing agent and building a custom AI system are different jobs. The scope, the assumptions and what is explicitly out of scope are all written down before work begins.
EIGHTY8