In June, AWS announced a $1 billion investment in a new organization built around a single job title: the forward-deployed engineer. Not a model. Not a chip. Thousands of people, embedded directly inside customer teams, with a mandate to compress AI deployments "from months to days."
A month earlier, Accenture and ServiceNow launched their own Forward Deployed Engineering program at Knowledge 2026. The stat they anchored on: 88% of enterprise AI initiatives never make it into scale production.
The forward-deployed engineer is the defining role of enterprise AI in 2026. And I think almost everyone is drawing the wrong conclusion from that.
Why this role exists
The FDE is a Palantir invention: an engineer who embeds with a customer, configures a complex product inside the customer's real environment (their data, their systems, their constraints), and stays accountable until it delivers value. Not a consultant who leaves recommendations. Not a support rep who answers questions. Someone who does the work.
The AI industry rediscovered the role because AI exposed a brutal gap. The demo is magic; production is where initiatives go to die. The model was never the problem. The missing layer is deployment: connecting the product to the customer's actual systems, configuring it for their actual use case, and getting their actual users to value.
That's why AWS is willing to spend a billion dollars on it, and why customers like the NFL credit FDEs with getting them "into production in just weeks." Deployment capacity is the binding constraint on the entire AI economy right now.
The math that doesn't work
Here's the problem: everyone is scaling this role with humans, and the economics only work at the very top of the market.
A widely-cited founder's playbook puts FDE compensation at $200K–$250K all-in at seed stage, rising to $450K–$550K+ at Series C. A single FDE embeds with one customer for 60–180 days. The standard advice is to hire your first one only after you've closed multiple pilots above roughly $50K ACV, because below that, the unit economics collapse.
Do the arithmetic. A human FDE serves a handful of accounts per year and costs as much as three product engineers. That works if you're Palantir selling eight-figure contracts. It works if you're AWS amortizing across the Fortune 500. It does not work for the tens of thousands of software companies whose customers pay $5K, $20K, or $50K a year — which is to say, almost all of them.
So the market is resolving the shortage two ways. The first is services: agencies now sell forward-deployed engineering by the hour. The second is a first wave of products aimed at the buyer's side of the problem: june.ai emerged from stealth this month with $20 million to automate enterprise software implementation, and lab0 is building "the open AI FDE" to deploy ServiceNow, SAP, and Workday in weeks instead of quarters. Both are real, both are well-aimed, and both automate the deployment of software an enterprise bought.
Which is worth pausing on, because the forward-deployed engineer was never the buyer's engineer. Palantir's FDEs deploy Palantir. The role, from its invention, is the vendor's engineer, embedded forward into the customer until the vendor's product delivers value. Automating the systems integrator is a real and valuable business. It just isn't, strictly, the FDE. The vendor's side of this category — an FDE for the software you sell — is still open.
Every software company needs the FDE function. Almost none can afford the FDE headcount. When a role is universally needed and economically impossible to staff, it doesn't stay a job. It becomes software.
What an AI forward-deployed engineer looks like
Decompose what an FDE actually does and something interesting appears: nearly every piece of the role has just become tractable for an agent — not because models got smarter, but because the action layer got standardized. MCP gave software companies a way to expose their product's capabilities as tools an agent can call, authenticated as a real user, with real permissions.
That changes what's automatable. Here's the role, piece by piece:
Pre-sales scoping. A real FDE joins early conversations, understands what the prospect needs, and shapes the deployment before the contract is signed. An AI FDE does this at the first touch — on your website, or when a prospect's own AI agent contacts your company through an agent gateway — asking the questions that matter and qualifying against your actual criteria.
Environment setup. A real FDE provisions and configures the product for this specific customer. An AI FDE provisions the account through your product's own endpoints: plan, settings, integrations, sample data, drawn from everything learned before signup, finished before the customer's first login. The empty trial — "here's your blank workspace, figure it out" — simply stops existing.
Hands-on configuration. A real FDE sits with users and does the work in the product. An AI FDE operates the product alongside every user with native in-app actions, shows each step as it goes, and calls the product's tools as the signed-in user when that's the more reliable path.
Staying accountable. A real FDE doesn't disappear after go-live. Neither does the agent: customers keep asking, and it keeps doing — configuration changes, reports, troubleshooting — escalating to your team only when judgment is required.
This isn't hypothetical. Starshipit is running this model now: within five weeks of rolling the agent into their app, 23% of new users were getting hands-on help, the median user's first session with the agent came 15 minutes after signup, and users who engaged during their first month were 2.6× more likely to still be active the next month.
What stays human
Forward-deployed engineers who write custom integration code against undocumented mainframes are not being replaced by an agent this year, and it would be dishonest to claim otherwise. The AI FDE works through the product's surface and the tools the vendor chooses to expose. Humans keep the things that deserve humans: policy, architecture, the relationships that matter, and the exceptions that need judgment.
But that's exactly how every leverage shift in software has worked. The human FDE becomes the person who defines what the agent is allowed to do and handles the 1% of deployments that are genuinely novel. The agent does the other hundred thousand.
Who feels this first
Two kinds of companies are living this problem most acutely.
AI companies — the very companies hiring FDEs off the comp curve above. Their products are complex, their deployments are the bottleneck, and they're MCP-native, which means the action layer drops in with almost no lift.
Sales-led software companies with complex products. These companies never had a self-serve motion because their product genuinely requires deployment work. An AI FDE doesn't just make their trials better — it manufactures a self-serve motion that could not previously exist. Every prospect gets the white-glove deployment that used to be reserved for the biggest deals. We've written before about why most B2B free trials fail in the first 48 hours; for these companies, the trial couldn't even fail, because it couldn't exist.
The claim
Here's the prediction this whole essay exists to make: within a few years, every serious software company will have forward-deployed engineers, in the sense that every serious software company will have an agent that scopes, provisions, configures, and operates its product for every customer, from the first conversation to the ten-thousandth task.
The role that AWS is spending a billion dollars to staff with humans is, for everyone who isn't AWS, going to be software.
That's what we're building at Aimdoc: an AI forward-deployed engineer for every one of your customers — one agent, embedded from the moment a prospect (or their AI) first asks about your product to every task they ask of it afterward. Others are deploying the software you bought. We're the FDE for the software you sell. Deployment, in our view, was never a headcount problem. It was always a leverage problem. And leverage is what software is for.
Want an AI forward-deployed engineer for every one of your customers? Explore Aimdoc Application, the Agent Gateway, or book a demo.