India Today reported on 10 July 2026 that forward deployed engineering can pay up to a million dollars a year; The Economic Times reported the same role at $200,000 on 3 June 2026, and neither says whether the figure is base pay or a whole package.
The numbers that survive a screenshot are narrower. Anthropic does not post a title called Forward Deployed Engineer at all. Its own posting for Head of Applied AI Architecture, Beneficial Deployments names the role once, in a requirements bullet, and states $315,000 to $380,000. The nearest posted title, Applied AI Architect, states $240,000 to $315,000. Both checked 25 July 2026.
The Wall Street Journal's headline on 19 July 2026 was "The hottest job in tech isn't very glamorous." Unglamorous is the job. You fly to somebody else's office, log into a network you do not control, and make their mess produce something. AWS has put a billion dollars into embedding engineers like this inside its customers. TCS is building a team of 8,900. Google is hiring hundreds.
The thing that stops you is not the model. It is a security reviewer who will not sign your deployment.
Your agent works. The pilot has been parked for two months while somebody decides whether the code you demoed can be allowed near their network.
This book builds the document you hand that person, and it has a name. The Approval Pack. Seven files you write, two tables your own programs generate, and verify.sh.
You stand up a self-hosted Claude apps gateway on a private address behind your customer's ingress, complete a device-grant login against the identity provider they hand you, serve per-group policy a developer cannot override from their own laptop, and route inference through the customer's own Amazon Bedrock account on pinned inference profiles, so no laptop on the engagement holds a cloud credential.