AI can write code quickly. Turning that code into a system that actually runs inside a company is a different job. That job now has a name: forward deployed engineer, or FDE. It is a software developer sent inside a customer’s organization to sit with their people and wire AI into real systems until it works.
Anthropic, a company often associated with the idea that AI will write the code and reduce the need for developers, has committed $100 million to train 10,000 of these engineers by the end of 2027. Big banks, a drug company, and some of the largest consulting firms are already sending their strongest engineers into that training.
What a Forward Deployed Engineer Actually Does
An FDE is not hired to produce a demo. A company places the engineer with the customer. The engineer learns the business, reads the code, connects one system to another, and owns the result when something breaks at 2 a.m.
That gap shows up often. A team can show an AI demo for invoice approval by lunch. The demo is not connected to the systems the company already uses. A vibe-coded prototype does not help if it is not integrated. Someone still has to decide what must happen, what must connect to what, and who will make it work.
Banks, life science companies, manufacturers, and large law firms are bringing in more technical talent for this reason. When the data is bank records or patient charts, it is not handed to just anyone or simply pushed into the cloud. The work needs a developer on site who knows the systems and the rules.
Palantir made the model well known. For nearly 25 years it has sent software developers on site, into the data, to connect systems and build them where the customer works. Frontier labs now want a piece of that same motion.
How Fast the Role Is Growing
Job posts for this role on Indeed grew 729%. Measured against January 2025, the increase is more than 5,000%. Separate reporting puts growth at 729% over the last 12 months, and other coverage puts postings up more than 700% last year.
Pay on Indeed runs from about $170,000 to $200,000 a year. One job-market map lists 1,300 live postings, 565 companies hiring, and an average median salary of $193,000. About 9% of openings sit at frontier labs and Palantir. Five of the largest AI companies are hiring for more than 9,000 of these roles. Someone who rents these engineers out has said the growth is likely to continue.
A company’s first forward deployed engineer requirement is described as an early public signal that it is shifting away from a purely product-led approach. The role is also more global than its reputation. At well-funded labs, matched base roles can carry equity in the 55% to 85% range by Series B median. Vertical experience integrating AI is said to pay more than seniority alone, and more than a wide range of general software development experience.
Despite layoff headlines, this is presented as one role that still looks secure, and as one of the most important functions in AI rollouts.
Anthropic’s $100 Million Training Push
Anthropic is backing Cloud Frontier Academy, also referred to in the launch language as Cloud Frontier Academy, with a $100 million commitment aimed at the enterprise AI talent gap. The goal is 10,000 FDEs trained by the end of 2027.
The first cohort includes engineers from Accenture, Bain, and other large firms. Morgan Stanley, Commonwealth Bank of Australia, and Novo Nordisk are in that first class. Companies nominate their strongest people. Those engineers go through an in-person simulated enterprise deployment, then spend 12 weeks on a real cloud deployment inside their own organization.
Golden Satch has opened a new office for 125 more AI and cloud engineers. Its CEO said people remain the firm’s greatest assets.
Why the Model Is Not the Hard Part
AI writing code quickly is not the disputed point. A flashy demo is still a long way from a working system deployed into an existing ecosystem. A bank is not going to rewrite its banking system overnight. A healthcare organization is not going to throw away everything it already runs. The work is integration.
One cited figure is that 95% of generative AI pilots still return nothing to the profit and loss statement. The variable described as doubling the odds of reaching production is not a better model. It is embedding the engineer who deploys the system.
Deployment itself is only about 20% of an AI system’s lifetime cost. The remaining 80% is paid later, which is where lock-in sits. Knowing how to deploy an on-premises system is framed as an advantage in that job market. Frontier labs are pitching operating companies on deploy engineers who deploy their own stack.
Mark Cuban has compared the moment to the early PC era. In the early 1990s, wealth was built by helping businesses deploy the PC. Companies already have PCs, systems, and SaaS. What is new is a wave of AI demos that still have to be connected to those existing systems.
Andreessen Horowitz put a similar idea in 2025: companies buying AI are like a grandmother with a new phone. She wants to use it, but she needs someone to set it up. In another piece, the firm framed the forward deployed engineer as trading margin for a moat, and called the role the hottest job in startups. The person who makes the promise is the same person who has to make it true.
Simon Willison has said that the more time he spends with coding agents, the more convinced he is that they make software engineering harder. Amazing work is possible, but unlocking their full potential takes extraordinary discipline and knowledge. Even then, someone still has to deploy the result, connect it to legacy systems, and make AI able to use legacy data.
Local Systems Versus Vendor Lock-In
Anthropic’s training spend is also read as a path to vendor lock-in: more organizations pumping data into Anthropic. That approach is called a bad idea for customers who do not want their systems tied to one vendor’s cloud or pricing.
The alternative argued here is local AI. AI should be infrastructure a company owns, sitting on its own machines, serving its own code, and answering only to it, rather than a subscription it rents. OpenMonoAgent, at openmonoagent.ai, is described as a terminal-native AI coding agent that runs entirely on local models, with zero API costs, zero telemetry, and full ownership. The claim is unlimited tokens and a stack the company controls. A forward deployed engineer who is not talking in those terms, the argument goes, is the wrong fit.
What This Means for Software Developers
Software developers talking about the end of coding jobs are looking at the wrong signal if this hiring wave holds. The growth figures cited here, from more than 700% to 729% and, versus January 2025, more than 5,000%, point to demand for people who can integrate AI, not only generate it.
Startuphakk, led by Spencer Thomason, builds custom software and treats this integration work as the core of the job. Thomason has 25 years in software development and a decade of executive leadership as a fractional CTO, including time with organizations such as GoDaddy, SRP, and Wells Fargo. The team’s work covers database architecture, API design, system integration, and infrastructure, with AI designed into the system when it belongs there, running in the customer’s environment rather than bolted on as a wrapper around someone else’s API.
The closing claim is that most companies do not have a technology problem. They have a leadership problem, paid for in missed deadlines, failed integrations, and AI investments that deliver nothing. The businesses described as winning are the ones built on solid engineering that treat AI as infrastructure they own, control, and integrate into software that actually works.
Developers who want to move toward this role are pointed at learning how to integrate AI and at building local systems. Companies that need the work done are pointed at engineers who will get into the existing systems and own the result after the demo is over.
Conclusion
The hottest coding job in this account is not the person replaced by AI. It is the person who makes the AI work. Anthropic’s $100 million plan to train 10,000 engineers by the end of 2027, the surge in forward deployed engineer postings, and pay running from about $170,000 to $200,000 all point the same way. Demos are easy. Production is not.
A bank, a life science company, a manufacturer, or a law firm still needs someone who can read the code, learn the business, connect old systems to new ones, and own the break at 2 a.m. Palantir built that model over nearly 25 years. Frontier labs and large enterprises are now hiring for it. The cited odds of a pilot reaching production rise when that engineer is embedded, not when a better model is chosen.
For developers, the opening is learning how to integrate AI and how to deploy it, including on systems a company owns rather than rents. For companies, the opening is the same person who makes the promise also being the person who makes it true. AI can write code fast. Someone still has to put it into place.
