Two years ago, we made a decision.
We would prepare our engineers for the work businesses increasingly need: bringing AI into their everyday operations. Not as a slide in a pitch deck, but as something that runs on Monday morning and still runs on Friday night.
We started with ourselves. With our team. With our own product.
What a forward-deployed engineer actually does
For the past two years we have been retraining our engineers as forward-deployed engineers. The title sounds grander than the work. In practice it means an engineer who sits next to the business problem, not two tickets away from it, and who can take an AI model from "interesting demo" to "part of how this company operates".
That required learning a set of skills nobody taught us at university:
- Prompting with precision — knowing the difference between a request that sounds clear to a human and one that is actually unambiguous to a model.
- Giving AI the right context — deciding what the model needs to see, what it must never see, and how to keep that boundary when the system grows.
- Orchestrating agents — breaking a workflow into steps, letting agents handle the parts they are good at, and keeping humans at the points where judgment matters.
- Checking their work — building the review loop, the tests and the guardrails, because a model that is right 95% of the time is a liability in the remaining 5% unless someone designed for it.
None of these replace classical engineering. They sit on top of it. An engineer who cannot read a legacy codebase will not get much from an agent that writes into it.
We started with our own team
The first client was us.
We rebuilt how our own team works around agents. The assistant that runs our internal chat today is a forward-deployed engineer's project: it reads the conversation, keeps track of tasks and reminders, edits and publishes our landing pages, and performs operations on our servers — every change going through a confirmation step before it touches anything real. It was built, tested and shipped by the same people we now send to clients.
That matters for one reason: we made our mistakes on ourselves first. We learned where agents drift, where context runs out, where "it worked on my machine" turns into "it worked on my prompt". By the time we brought this to a client's operations, the painful lessons were already behind us.
Then with the businesses we work with
Over these two years we have brought AI into the everyday operations of the businesses we work with — not as a separate "AI project", but inside the processes that already existed: support, internal tooling, data handling, code maintenance.
The pattern that keeps repeating is simple. The hard part is never the model. The hard part is the context: understanding how the business actually works, which decisions can be delegated to a system and which cannot, and who is accountable when something goes wrong. That is exactly the work a forward-deployed engineer is trained for.
Why this matters now
Businesses are no longer asking whether to use AI. They are asking who will make it work inside their operations without breaking what already works. That is an engineering question, and it needs engineers who understand both sides: the model and the business.
These skills are not a trend we picked up last quarter. They are two years of retraining, of shipping to ourselves first, and of being honest about what AI does well and where it still needs a human in the loop.
We are ready to bring that to your operations. And if you are an engineer who wants to work this way, we would like to hear from you too.