FORWARD DEPLOYED
ENGINEER

AI is an intelligence layer, not an app. The most successful AI companies in the world (OpenAI, Anthropic, Palantir) use Forward Deployed Engineering to implement it. They embed engineers inside client operations because SaaS doesn't work. The intelligence layer requires someone who understands the client's reality, maps how work actually moves, and builds systems that stick.

Three years ago, Scelta adopted a boots-on-the-ground implementation strategy because SaaS didn't work for our clients. We didn't know what FDE was. We just knew that showing up, interviewing teams, extracting undocumented knowledge, and building systems inside client operations was the only thing that delivered results.

We arrived at the same methodology independently. Now the discipline has a name.

OpenAI wants 5 to 7 years of experience, production LLM deployments, and a proven track record. That's a high bar.

Understanding the thesis doesn't require seven years. It requires paying attention.

You've been building with these tools. You've hit the wall where it's functional but unreliable. You've seen that the demo isn't the product. You understand that AI at the application layer, where it touches real operations, almost always needs embedded engineering to make it work.

You just haven't had a path in. The discipline just got named. The big companies want senior people.

This is the path.

We have an R&D project starting on April 1st that requires FDE positions.

The intelligence layer splits into two lanes. Some use cases work with off-the-shelf tools: NotebookLM, Claude, Vercel. Deploy the package, train the client, move on. But construction and industrial use cases often don't. They're functional but unreliable. They need bounded validation, failure-mode logging, test sets, evidence trails.

We're building the routing system. Classify the request. Send it to Production if a validated package exists. Send it to R&D if it needs validation. Refuse or escalate if uncertainty is too high.

You'll work with clients to surface use cases. Route them to the correct lane. Deploy what works. Document what doesn't. Build the test sets. Promote validated patterns into deployment packages.

You'll help write the playbook, not follow one.

You're organized. The work is structured interviews, clean documentation, proper folders. You extract information and build repositories that last.

You can talk to anyone. Finance teams, operations leads, the person who's been there 30 years and knows everything.

You want to build something that has compounding, meaningful impact.

Bootstrapped. Customer-funded. No outside capital telling us what to build.

Boots on the ground. We work inside client operations. We show up.

We run hill sprints. We track recovery. We believe healthy people build healthy businesses.

The work speaks. Sophisticated operators find us. We're not loud about it.