In roughly 90 days, OpenAI, Microsoft, Amazon, and Google all built dedicated deployment organizations, committing more than $7.5 billion in total. Salesforce committed to hiring 1,000 forward-deployed engineers. One analysis of more than 1,000 job postings found the role grew 1,165% year over year. The title is twenty years old. The scramble is new.
Something unusual happened in the first half of 2026. Four of the largest technology companies in the world, competitors in almost every other respect, independently reached the same conclusion in a single quarter: selling AI is not the hard part, and neither is building it. Getting it to work inside a specific company's messy reality is the hard part, and it requires a kind of engineer most organizations do not employ.
OpenAI announced its Deployment Company on May 11, 2026, with a $4 billion initial investment, 19 partner firms, and roughly 150 forward-deployed engineers and deployment specialists acquired through its purchase of the London applied-AI firm Tomoro. On July 2, TechCrunch reported that Microsoft had launched its Frontier Company with a $2.5 billion commitment and 6,000 industry and engineering experts, two days after Amazon launched its own $1 billion deployment venture. Google Cloud CEO Thomas Kurian announced plans to hire hundreds of people.
For anyone responsible for staffing an AI initiative, this is a signal rather than industry news. The companies with the best models on earth concluded that the models were not enough. This article covers what the role actually is, why it exploded, what the hiring data shows, and what to do if you cannot realistically hire one.
What Is a Forward Deployed Engineer?
A forward-deployed engineer is a software engineer who embeds directly within a customer's organization to build production systems in that customer's environment and then remains accountable for adoption. Palantir, which originated the role, describes the inversion cleanly: a conventional software engineer creates a single capability used by many customers, while a forward-deployed engineer enables many capabilities for a single customer.
The role is not new. Palantir's FY2020 annual report described forward-deployed engineers who "have traveled to bases in Afghanistan and factories in the industrial Midwest to deploy our platforms." What changed is that a role invented to deploy one company's unusually complex software became the answer to a problem every AI vendor now shares.
The distinction that matters most in a staffing conversation is ownership. A solutions engineer supports the sale and hands off at the point of signature. A consultant diagnoses a workflow and recommends changes to it. A forward-deployed engineer builds the system that changes it and stays responsible once it is live, including the exceptions, evaluation harnesses, controls, and adoption problems that only appear in production. OpenAI describes its own engineers as working "inside the organization to design, build, test, and deploy production systems, connecting OpenAI models to the customer's data, tools, controls, and business processes."
The cleanest empirical evidence that this is not a sales role in disguise comes from job posting data. In an analysis of more than 1,000 forward-deployed engineer postings, 70% included equity and exactly 0% were quota-carrying.
Why Did the Biggest AI Companies Build Deployment Arms Within Ninety Days?
Because the bottleneck moved. When model capability was scarce, whoever had the best model won. Model capability is no longer scarce in any way that matters to a typical enterprise buyer, and the constraint shifted downstream to whether anyone can wire that capability into a specific company's data, permissions, and workflows and get people to actually use it.
Salesforce's account of its own model is instructive because it describes a delivery structure rather than a job title: forward-deployed engineers deployed either solo or in three-person pods, typically one deployment strategist and two engineers, over roughly three months per client. That is not a hiring plan. That is a professional services organization described in engineering terms.
Microsoft's Commercial Business CEO went out of his way to resist the label, saying the Frontier Company "goes beyond what has been labeled as Forward-Deployed Engineering" and would be "the largest, most capable, outcome-driven engineering organization." The pushback is telling. Four companies built the same thing at the same time, and at least one would rather not admit it looks exactly like consulting.
For buyers, the implication is uncomfortable and worth stating plainly. If the companies that build frontier models believe their software cannot land without embedded engineers, the assumption that your team can buy a license and self-serve its way to production deserves a second look.
What Does the Hiring Data Actually Show?
The most rigorous public dataset comes from an analysis of more than 1,000 forward-deployed engineer postings from the jobs-data provider Revealera, comparing January through October 2025 with the same period in 2024. Postings grew 1,165% year over year. Median salary was $173,816. The skills profile skewed heavily toward engineering: Python appeared in 66% of postings, TypeScript in 35%, AI agents in 35%, AWS in 32%, and LLMs in 31%.
One finding runs counter to the prevailing narrative. Fifty-eight percent of these roles sat at companies with 11 to 200 employees. Despite the volume of commentary about large enterprises building internal deployment teams, hiring is concentrated at AI vendors and startups deploying to customers, not at ordinary enterprises staffing their own teams. Claims to the contrary come primarily from executive search firms, whose business benefits directly from that belief. Treat them as estimates of interest rather than as measured labor data.
A second analysis of roughly 1,000 live postings in the first half of 2026 found core engineering requirements in more than 95% and AI-specific requirements in more than 80%, but the fastest-growing requirements were the customer-facing ones: requirements discovery, stakeholder management, and problem decomposition, present in more than 70% and rising. The scarcity is not in the engineering. It is in engineers who can run discovery in a room full of operators and then go build the thing.
Can You Actually Hire One?
Probably not at the price you have in mind, and the posted salary ranges tell the story better than any commentary.
Anthropic posted a forward-deployed engineer role on the federal civilian side in New York, with a salary range of $280,000 to $320,000, with responsibilities including working within customer systems to build production applications, delivering technical artifacts such as MCP servers and sub-agents, and traveling 25% to 50% of the time to customer sites. Palantir, which invented the role, posts its forward-deployed software engineer role at $135,000 to $200,000 plus equity, requiring as little as one year of post-college experience. Google Cloud's bands range from roughly $127,000 for an applied forward-deployed engineer to $265,000 at the most senior level, before bonus and equity.
That spread, roughly two-to-one for the same nominal title, is what a scarcity market looks like before anyone agrees on what the job is. Executive search commentary in mid-2026 estimated that around 2,000 US engineers have a genuine record of repeatedly deploying enterprise AI into production, out of perhaps 17,000 who hold the title. Those are one search firm's estimates rather than a labor statistic, and should be read as directional. The direction is not in dispute: you are competing for this profile against organizations that just committed billions to hiring it.
There is also a quality distinction buyers consistently miss. As the CEO of one applied-AI firm put it, many forward-deployed engineers are well equipped to help you roll a coding tool out to your workforce, but very few can build your flagship AI product feature. Those are different hires at different prices, and conflating them is how a search stalls for six months.
What Should You Do If You Cannot Hire One?
Separate the two jobs before you write a requisition. Rolling out tooling and driving adoption is one role. Building a differentiated AI capability into your product or core workflow is another. Most companies write one job description covering both, price it at the lower end, and then cannot understand why no qualified candidates apply.
Do not treat it as a services role you can outsource and forget. The reason the model works is continued ownership through production. An engagement that ends at handoff reproduces exactly the failure the role exists to solve. Whoever builds it needs to still be there when the workflow changes in month four.
Consider where the deployment talent actually is. When Google Cloud opened its initial wave of forward deployed engineer roles, the postings spanned the United States, India, Brazil, Australia, Mexico, Singapore, South Korea, and Canada. The companies competing most for this profile are not treating it as a role that has to be based in San Francisco. Latin America appeared in that first wave for the same reasons it appears in our clients' plans: real engineering depth and overlapping working hours, which matters enormously for a role defined by sitting with operators while they work. Our guide to AI staffing models covers how those engagement structures compare.
Buy the pattern, not just the person. What makes deployment work is a repeatable structure, not a heroic individual. A small embedded pod with an engineer who can build, someone who can run discovery with your operators, and clear accountability for the outcome is the thing to replicate. That structure is what an AI dedicated team is, and it is how tailored AI agents get from a working demonstration to something your team relies on. Our AI credit scoring and healthcare AI work both ran on that model.
Be honest about whether you need this at all. One practitioner's warning is worth repeating: forward-deployed engineering is, by definition, an upmarket motion, and you should not do it if your product's end shape is self-serve. The same applies internally. If your use case is an off-the-shelf tool solving a standard problem, you need a good rollout, not an embedded engineer.
Common Questions About Forward Deployed Engineers
What does a forward-deployed engineer do?
A forward-deployed engineer embeds with a customer or business unit, runs discovery with the people who actually perform a workflow, builds a production system around it in the customer's own environment, and remains accountable through adoption. Unlike a solutions engineer, the role continues past the sale. Unlike a consultant, the deliverable is working software rather than a recommendation.
How is a forward-deployed engineer different from a solutions engineer or consultant?
Ownership through production. Solutions engineers support technical sales, demonstrations, and configuration, then hand off. Consultants diagnose and recommend. A forward-deployed engineer builds the system that implements the change and stays responsible once it goes live. Job-posting data reinforces the distinction: none of the postings analyzed were quota-carrying.
How much does a forward-deployed engineer cost?
Posted US ranges in 2026 span roughly $127,000 to $320,000 in base salary, depending on the company and seniority, with equity often making up a large share of total compensation at AI labs. One analysis of over 1,000 postings put the median at $173,816. The widespread variation reflects a market that has not yet standardized the title's meaning.
Should a mid-sized company hire its own forward-deployed engineers?
Usually not as a first move. The hiring data shows that the role is concentrated among AI vendors and startups deploying to customers, rather than among ordinary enterprises staffing internally. For most companies, the practical path is an embedded team that provides the same structure, discovery, build, and ownership through adoption without competing for a scarce and expensive title against organizations that just committed billions to it.
If your AI initiative is stalled somewhere between a working demonstration and something your team actually uses, the missing piece is usually this role, not a better model. Talk to the Golabs team about putting an embedded engineering pod on the workflow that matters most.