AI Forward Deployed Engineer
An AI forward deployed engineer for the work that decides whether a pilot ever ships: discovery, system archaeology, field mapping, verification against the real rules, and the change requests that never stop after go-live — running on your infrastructure, under your credentials, inside your review process.
The category, as covered by
01 — The gap
A model that answers well in a sandbox tells you almost nothing, because production failure looks nothing like a wrong answer. It looks like an integration surface the model cannot reach. Two systems whose fields share a name but not a meaning. An approval branch the demo never exercised. A permission boundary that works only for the engineer who built it. An upstream schema that quietly moved in June, three months after go-live.
The role invented to close that distance is the forward deployed engineer — someone who sits inside the business, learns how it actually operates, translates that into something buildable, ships it, and owns the outcome afterwards. It works. It is also headcount: every enterprise is hiring for the same scarce people, capacity grows one engineer at a time, and the context walks out the door with them.
“The FDE model is rapidly gaining momentum as organisations look to fast-track AI integration and adoption by embedding engineering talent directly with client teams.”
“The forward deployed engineer fills that gap, ensuring that AI progresses with customers rather than drifting away from their needs.”
“AI companies are hiring forward-deployed engineers to work directly with customers and tailor AI tools to real-world use cases.”
02 — The role
Six steps, run in a loop. It is not a mystery — it is just relentless, and it is where the headcount goes.
How the work actually happens, not how the requirements document says it happens.
Turn business pain into a specific, solvable problem worth pointing a model at.
Decide what the system does, what data it touches, and how anyone will know it worked.
Fit the existing stack — the data, the permissions, the network, the compliance review.
Ship into a live environment where real people now depend on it.
Watch it, fix what fails, and keep changing it as the business changes underneath.
Orfyn does not replace the judgement in steps one and two. It removes the reason those steps stall — the months of execution behind them.
03 — The software
Orfyn is a self-hosted platform for running and orchestrating coding agents against your real systems. Five things make it usable on work that matters.
Run many agents at once instead of one engineer in sequence. Each gets its own isolated checkout, so parallel work does not collide.
Nothing lands on main because an agent decided it should. Work happens in isolation and arrives as a diff you review.
The parts that must be repeatable are scripts, not prompts. Anything you can do in the interface, you can run unattended.
The daemon runs on your hardware — a laptop, a homelab, a server in your own VPC. Your code does not leave it, and we do not sit in the path of your model calls.
The second change should cost less than the first. Context lives in the workspace, so the system that shipped is still the system you edit.
04 — Integrations
No connector catalogue to wait on. Agents operate the systems your teams already run — from the ERP to the CI pipeline — using the same interfaces your engineers use, under the same credentials.
Systems of record & GTM
DevOps & infrastructure
Agent harnesses it runs natively
UpfynAI
Claude Code
Codex
GitHub Copilot
Cursor
Gemini
OpenCode
Cline
Goose
Qwen Code
Kimi
Amp
Factory Droid
Mistral
PoolsideAll product names and logos are property of their respective owners, shown to indicate systems Orfyn agents can operate — not partnership, certification, or endorsement.
05 — Where it runs
Most AI vendors ask you to move your work to them. Orfyn is the other shape: the daemon runs on your machine, holds your keys, and talks to your systems from inside your network. There is no vendor session in the middle to audit, because there is no vendor session.
Laptop, homelab, or a server in your own VPC. macOS, Windows, and Linux are all first-class.
Bring your own provider accounts. Orfyn does not proxy model calls and adds no margin on inference.
Agents reach what the machine and the credentials they run under can reach. Nothing is escalated on your behalf.
Multi-provider by design. The best model changes every few months; the workflow around it shouldn't have to.
SOC 2
built against control objectives
ISO 27001
aligned control design
UK/EU GDPR
privacy by architecture
DPDP Act 2023
India-ready
E2E encryption
relay carries ciphertext only
Self-hosted
your keys, your network
Certification status, control evidence, and completed questionnaire responses are available under NDA — ask for the security package in the meeting.
06 — Go deeper
One deployment carried end to end — discovery, system exploration, build, verification, and the part after go-live.
The same loop under different constraints: enterprise SaaS, financial services, healthcare, manufacturing, public sector.
Your infrastructure, your keys, your boundaries — and why that is architecture rather than a policy document.
07 — Who it's for
The same daemon, the same isolation, the same review-before-merge — scaled to the size of the team running it.
More surface area than headcount.
Enterprise problems without enterprise headcount.
Capacity without the security fight.
08 — Feedback
Published only with written permission — every quote on this wall will carry a name, a role, and something checkable.
Awaiting publication
CTO · Early deployment
This slot is reserved for permissioned feedback from an early production deployment.
Awaiting publication
Head of Platform Engineering · Early deployment
Operator quotes are published only with written sign-off — names, roles, and results included.
Awaiting publication
VP Engineering · Early deployment
First-cohort feedback lands here once the deployment it describes is live and verified.
Awaiting publication
Founder · Early deployment
No invented praise. Every quote that appears on this wall will be attributable and checkable.
Awaiting publication
Director of Data · Early deployment
Reserved for the team that ships the first regulated-environment rollout on the platform.
Awaiting publication
Engineering Manager · Early deployment
Until then: bring one workflow to the meeting and judge the work directly instead.
These are reserved slots, not customers. Ask in the meeting for reference calls while the wall fills in.
09 — Questions
Software. The forward-deployed-engineer model works because someone technical owns the outcome inside your business — but delivered as consulting, it costs what consulting costs and stops when the engagement stops. Orfyn is the platform that does that work, running on your side.
Nowhere. The daemon runs on hardware you control and talks to your systems from inside your network. You supply your own model provider keys, so inference goes directly from your infrastructure to the provider you chose. There is no telemetry on your code.
Whichever ones you want. Orfyn is multi-provider by design and runs the agent harnesses your engineers already use — Claude Code, Codex, Copilot, OpenCode, and others, alongside custom providers. You can switch without rebuilding anything around them.
Isolation, not good intentions. Agents work in separate git worktrees with their own allocated service ports, so parallel work cannot collide and nothing touches your main checkout. Changes reach you as a diff to review.
You bring one workflow you are actually trying to ship. We go through what it touches, where it currently stalls, and what running it on Orfyn would look like. If it isn't a fit, we will say so in the meeting rather than after it.
The fastest way to find out whether this fits is to walk through something real. No deck, no download, no trial to configure.