Forward deployed engineers, carrying a factory.

Because the req is competing for a candidate profile that barely existed two years ago, and every AI company on the market is bidding for the same people. The role asks for staff-level engineering, customer-facing judgement and startup-speed execution in one person, so the search runs long, the offer runs high, and the ramp starts only after all of that clears. Contracting changes what you are buying. You are not buying a person you then have to make effective, you are buying an engineer who has done the install before, arrives with the tooling already built, and starts writing production code in your repo in the first weeks. Hiring is still the right end state when the work is permanent, and nothing here blocks it. Several of our engagements exist precisely so a team can keep shipping while the req stays open.

Staff augmentation sells you hours against a ticket queue. The contractor takes the tickets your team writes, works the way your team already works, and when the contract ends the only thing left is the code they wrote. A forward deployed engineer owns an outcome instead of a queue, and arrives carrying a system. They install the delivery infrastructure while they ship: context committed to the repo, skills maintained per repo, guards that decide whether agent output is mergeable, telemetry on every run. That is why we call every embedded engineer a soft factory install. The test is what remains after they leave. With staff augmentation it is a pile of merged tickets. With this it is a line that keeps running, and a team that knows how to operate it.

Week one is production code, not discovery theatre. The engineer picks up a real ticket from your backlog, ships it, and uses that pass to see where the friction actually is: which conventions are undocumented, which review comments repeat, where the design system is a suggestion rather than a rule. What they learn becomes committed context, not a report. By the end of the first month you have merged features from them, an AGENTS.md and CLAUDE.md pair your other AI tools read, and usually the first guard in CI that stops a class of review comment from ever being written again. If you want an artifact, ask for the list of guards. It is a better read than a status deck.

Mid-senior to senior only. Three years minimum, typically five or more, and the bar is deliberately two-sided: production-shipping discipline and AI-native workflow. We do not place juniors, and we do not place engineers who are fluent with a coding agent but have never carried a feature through review, staging and an incident. You interview whoever you want, on whatever loop you already run for your own hires. If someone does not clear your bar, that is a signal about the match, not something to negotiate around.

Same engineer, on a monthly retainer with a six-month minimum, because the value compounds only if the person accumulates context in your codebase. Holidays and reserve duty are real and we plan around them rather than pretending otherwise. The structural answer is the same one we give about handover: because the engineer commits context, skills and guards to your repo as they work, the knowledge is not held in one head. That is what makes a cover or a transition survivable, and it is the same property that makes the eventual exit survivable.

It stops being an embedded engineer and becomes a factory deployment. One FDE can install the practices in one team and prove them there. When three teams want the same thing, the constraint moves from the engineer to the system, and the right shape is a phased install with a defined outcome and a fixed price per stage rather than more retainers. That upgrade is the normal path, not an upsell we invented after the fact: the engineer was already installing the same infrastructure, just at the scale one person can carry. The Frontend Delivery Factory page is where that engagement is described end to end.

No, and the engagement is designed to end. Everything the engineer builds lives in your repository under your license: the context files, the skills, the CI guards, the eval harness, the traces. There is no Enpitech runtime in the path, no hosted service you have to keep paying for, and no configuration that only we can read. We use tools your team can keep using, whichever coding agent you already run. The honest version of the pitch is that we are trying to make ourselves unnecessary on a schedule, and the way you check we mean it is to ask, at the end, whether your own engineers can change a guard without calling us.