Every business that bought into AI over the last two years is now facing the same question: where is the payoff? A new report from executive search firm Christian & Timbers puts a number on why that question is so hard to answer. Out of everyone working in AI in the United States, the firm counts only about 2,000 people skilled enough to reliably turn AI spending into millions of dollars in actual profit.
These people are called forward deployed engineers. Instead of building a general AI product and selling it off the shelf, they sit inside a client's business and wire the AI into real workflows: sales, claims processing, document review, whatever the job requires.
The idea did not start recently. Palantir built this model more than a decade ago because its government clients could not simply describe what they needed on a slide deck, so someone had to sit with them and build it live.
What has changed is demand. At the start of the year, fewer than one in ten companies planned to hire for this role, mostly for small trial projects. By mid year, seven in ten companies wanted one, and the biggest consulting firms are trying to grow these teams tenfold.
The forecast is a 2,100 percent jump in demand for the role this year, landing on a talent pool that is nowhere near big enough to absorb it. The pay reflects the shortage.
A typical software engineer earns roughly 215,000 dollars a year in the US. A mid-level forward deployed engineer is closer to 385,000 dollars, and the most senior ones inside frontier AI labs are clearing a million dollars a year.
Companies are not just paying for coding skill here. They are paying for someone who can walk into a boardroom, understand an unfamiliar industry fast, and be trusted enough to touch sensitive internal data.
The urgency traces back to a hard number in the AI industry that will not go away: a widely cited MIT study found that 95 percent of corporate generative AI pilots fail to produce a measurable financial return. That gap between AI spending and AI payoff is exactly what these engineers are meant to close.
This fall, investors are expected to start rewarding companies that closed that gap and punishing those that did not. OpenAI and Anthropic feel this pressure directly, since both have already spent tens of billions of dollars building their models.
Both companies have launched their own deployment arms, OpenAI's Deployment Company and Anthropic's Ode joint venture, staffed with engineers whose only job is getting AI working inside big companies. That push matters more because cheaper, increasingly capable open source models out of China threaten the profits these labs need from selling access to their technology.
Many companies, however, are choosing to build this skill in-house rather than renting it from OpenAI, Anthropic, or a consulting firm. The logic is simple: handing your internal processes to the same company selling you the AI model means teaching a potential competitor how your business works.
The uncomfortable long-term truth, even from the people selling this talent, is that the role may not last. As AI gets better at automating itself, the specialists teaching companies how to use AI could eventually be replaced by the technology they deployed.