# On-prem, regulated, legacy: deploying AI where it's hardest

> The demo runs on a clean laptop. The job runs inside a bank's air-gapped network, a ministry's twenty-year-old database, a hospital's compliance regime. That gap is the whole role.

**Category:** The role
**Published:** 2026-07-16
**Author:** Pierre-Thomas Liger-Belair

Every impressive AI demo shares one thing: it runs in an environment the builder controls. The Forward Deployed Engineer works in the opposite condition — someone else's environment, and usually the least forgiving version of it.

On-prem. Many of the highest-value clients can't send data to a cloud API. The model runs inside their walls, air-gapped, on their hardware. That changes everything: no managed endpoints, no easy scaling, latency and memory budgets that are real constraints, and updates that ship through a change-control process, not a git push.

Regulated. In finance, health and government, "it works" isn't enough — you need auditability, access controls, data residency, and an answer for every decision the system makes. The eval set isn't a nice-to-have; it's the evidence you'll be asked to produce.

Legacy. The AI is the easy part. The hard part is the integration: a SOAP API from 2009, an identity system nobody understands, a database schema that encodes a decade of exceptions. Making a frontier model survive contact with that is the skill clients actually pay for.

This is why the role rewards temperament over brilliance. The engineer who stays calm in the mess — who treats the constraints as the problem, not an obstacle to the problem — is the one who ships.

---

Source: https://forward-deployed-engineer.tech/news/on-prem-regulated-legacy
Publication: Forward Deployed — the publication of record for Forward Deployed Engineering.
Editor: Pierre-Thomas Liger-Belair.
This markdown mirrors the HTML page at the URL above. Content is written and edited, not generated in bulk.