fd.engineering — Forward Deployed AI Engineers

The problem

Production agents are the hard part.

Anyone can demo an agent. Shipping one that does meaningful work — reliably, inside your environment, at the latency and cost your business needs — is different work. Evals that hold up under real traffic. Retrieval that survives the long tail. Failure modes the happy-path demo never surfaced. Integration with the systems the agent has to actually operate.

That's the gap. It's where most teams stall, and it's the work we do.

The role

What a forward deployed engineer actually does.

We're senior engineers who embed with your team, understand your domain fast, and ship production AI systems inside your environment. Not a slide deck. Not a Notion page. Working software, on your infrastructure, doing the thing you needed it to do.

We write the code. We debug the integrations. We sit in your standups. We leave when you can run it yourselves.

This is the Palantir model. The Anthropic-at-FIS model. The model OpenAI just put $4 billion behind. It works because the alternative — telling enterprises to figure it out from documentation — is the reason most AI initiatives stall at the pilot stage.

How we engage

Three ways to work with us.

Diagnostic 1–2 weeks

A working deployment plan, not a strategy memo. We scope the problem, map your stack, identify the build, and tell you what it costs. If we're not the right team, we'll say so. Often converts to a Deployment.

Deployment 6–12 weeks, fixed scope

We build and ship one specific capability end-to-end. A RAG system over your support corpus. A fine-tuned model for your domain. A custom-trained model when the frontier APIs don't fit. An agent that handles tier-1 tickets. A copilot embedded in your product. A vision pipeline running on the edge. We hand it off with documentation, evals, and a team that knows how to extend it.

Embedded 3-month minimum, monthly retainer

A senior engineer in your Slack, your standups, your codebase. For teams with an AI roadmap, not just a project. We build alongside your engineers, and we transfer everything we know on the way out.

Hiring a senior FDE means months of recruiting, six-figure comp packages, and ramp time before they ship. Ours start within days of signing. No recruiter fee, no equity, no ramp, clean exit when you're ready. We slot in alongside an existing FDE org or stand up the practice for you.

Where we're strongest

Five areas where we do our best work.

AI systems in production

Model integration, retrieval, agentic workflows, evals, the harness around the model. Where most teams stall.

Custom model training and fine-tuning

When the off-the-shelf model isn't enough — domain-specific fine-tunes, custom architectures, training pipelines that hold up in production.

RAG and retrieval that actually works

Past the demo, into the long tail of real documents and real queries.

Physical AI and robotics

Perception, world models, sim-to-real, human-in-the-loop autonomy. Rare expertise, hard to hire for.

Industrial and operational AI

Manufacturing, energy, logistics — places where the model has to live next to a PLC, not a Postgres.

If your problem isn't on this list and it's still AI, ask anyway. We'll tell you straight.

How we operate

Every engineer we deploy meets five bars.

Production AI experienceHands-on with LLMs, agents, and the infrastructure they actually run on — not familiarity, deployment.
Customer empathyWe understand why your workflow looks the way it does before proposing how to change it.
Comfort with ambiguityNo spec, no problem. We diagnose and scope simultaneously, with incomplete information and shifting requirements.
Executive communicationWe translate architecture decisions into language a CFO cares about, then walk back into the codebase and write the code.
System-level thinking across unfamiliar codebasesWe ship inside environments we didn't build, usually within days.

The industry calls this the FDE profile. We call it the floor.

Who's behind this

fd.engineering is led by Jason Fox.

25+ years building production systems, including five years as a forward deployed engineer at Microsoft — embedded with enterprise customers, shipping machine learning systems inside their environments before the term went mainstream. Currently the founder and lead engineer at fd.engineering and co-founder and CTO of Noumenal Labs, a Physical AI startup.

The work is hands-on. Jason writes the code. So does everyone we embed.

What we're not

Saying what we don't do is part of saying what we do.

We're not a consultancy. We don't sell decks, frameworks, or transformation roadmaps.
We're not an audit shop. If you want a maturity assessment, plenty of firms will sell you one.
We're not staff augmentation. We scope outcomes, not hours.
We're not generalists. We do AI systems — model integration, retrieval, agents, evaluation, deployment, and the production engineering around them.

When an engagement needs strategy work We bring in trusted partners who do that part well. You get one accountable team, not a procurement exercise across three vendors.

The honest part

Forward deployed engineering won't be a category forever.

Eventually the talent market catches up and your team builds the AI fluency to ship without us. That's fine — that's the goal. Every engagement we run ends with knowledge transfer and, when it makes sense, a recommendation for the full-time hire who replaces us.

We'd rather earn one great reference than a long retainer.

Contact

Let's talk.

Tell us what you're trying to build, what's in the way, and we'll get back within one business day. If we're not the right team, we probably know who is.

Or email jason@fd.engineering directly.