Forward Deployed Engineers for AI Companies — What Actually Matters
Why AI companies hire FDEs, what the role looks like when the product is agents, and how embedded delivery closes the last mile.
“We can demo it, but we can't get it into the customer's workflow.” That sentence is why AI companies are hiring Forward Deployed Engineers, a role Palantir made famous. An FDE sits close to the customer, maps the real workflow, ships software into it, and stays until the integration holds up under ordinary use.
When the product is agentic AI, the job gets sharper. The hard part is rarely the model. It's working out which work is agent-shaped, and then wiring agents into systems of record that were never designed to be called by software.
What Forward Deployed means for agent products
The loop I run:
- Discover: sit with the operators, watch the handoffs, name the failure modes nobody put in the RFP
- Design: pick the thinnest agent that can own one outcome end to end
- Ship: build the MCP tools and orchestration against the real APIs, real auth, and real data quirks
- Harden: logging, cost controls, human checkpoints, a rollback path
- Hand off: leave runbooks and an owner so it survives without you
Skip discovery and you ship a clever agent nobody opens. Skip hardening and you ship a liability. The discipline is doing both while the clock runs.
Why the pattern looks familiar to me
Eighteen years shipping production software, most of it inside someone else's business rather than my own. That maps onto the FDE job in four places:
- Embedded delivery: you win by being where the work happens, not by presenting at it
- Defined outcomes:agree what “done” means before you add the second agent or the third engineer
- Founder judgment:I've founded or co-founded eight SaaS products, so I know the difference between a demo that impresses investors and a system that survives month-end
- Agent stack fluency: Claude Code agents, custom MCP servers, and skills running in production against real customer data
How this sits next to fractional CTO work
Forward Deployed Engineering is how I show up for AI product companies that need someone shipping inside their customers' workflows. Fractional CTO is the model when an SMB needs senior architecture without a full-time hire. Same underlying work, different commercial wrapper.
If you're building an AI product and the bottleneck is the last mile into your customers, that's an FDE problem. If you need someone to own the agent roadmap part-time, that's fractional CTO. I do both through Red Leg Dev.
What good looks like
A Forward Deployed engagement should leave behind:
- One or two workflows with a measurable before and after
- Agents and MCP tools shaped like how the customer actually works
- Enough observability that failures are boring instead of mysterious
- A handoff their team can run without you on every call
That's the bar. AI companies that can't close the last mile don't keep the customer, however good the model underneath it is.