Forward-deployed engineers (FDEs) are software engineers a vendor embeds directly inside a customer’s organization to make AI actually work in production — scoping the messy real work, wiring it into legacy systems, and staying until something ships. In the span of three days at the start of July 2026, Amazon and Microsoft committed a combined $3.5 billion to building the largest FDE armies in their history. The tell buried in that number: the companies selling autonomous AI agents are simultaneously hiring thousands of humans to make those agents land — because the hard part was never the model, it was coordination. Lova is the chat-first AI project management product where AI agents work as first-class teammates on a shared board — claiming tasks, posting evidence, and moving cards only through verifiable status, right alongside human teammates — which is the coordination surface an FDE otherwise has to rebuild by hand inside every customer.
The wave broke fast. On June 30, 2026, AWS said it would put $1 billion into a Forward Deployed Engineering organization that embeds pods of engineers — working alongside AI agents — inside customer teams to co-build and deploy agentic AI in a matter of weeks. Two days later, on July 2, Microsoft answered with a bigger bet: a new subsidiary called Frontier Company, funded at $2.5 billion with roughly 6,000 employees who go sit inside customer businesses. When two of the biggest AI sellers on earth pour billions into humans in the same week, the interesting question is not whether agents work. It is why deployment keeps stalling — and the answer is the same one the data has been quietly repeating all year.
Key takeaways
- Between June 30 and July 2, 2026, AWS committed $1 billion and Microsoft $2.5 billion (about 6,000 people) to forward-deployed engineering — armies of humans embedded in customers to make AI ship.
- Demand for the role has gone vertical. Job postings for forward-deployed engineers rose from roughly 643 in April 2025 to 5,330 in April 2026 — about a 729% jump — per data reported by Business Insider.
- Two frontier AI labs reportedly stood up their own FDE joint ventures valued at $4 billion and $1.5 billion, per TechCrunch — the labs building the models are also renting out the humans to deploy them.
- Microsoft’s own 2026 Work Trend Index found that organizational factors explain 67% of AI’s reported impact versus 32% for individual capability — coordination matters more than twice as much as the model, even as agent use grew 15x year over year.
- The FDE boom is the market pricing a known gap: MIT found 95% of enterprise generative AI pilots deliver no measurable return, and Gartner expects over 40% of agentic AI projects to be canceled by the end of 2027. Both are coordination failures, not capability ones.
What is a forward-deployed engineer, and why is everyone hiring them?
The role was popularized by Palantir over a decade ago, out of necessity: the software was powerful, but customers in government and defense could not simply plug it into fragmented data, legacy workflows, and high-stakes constraints. So Palantir embedded engineers directly inside customer teams — people who would write production code on-site, untangle the data pipelines, and translate raw capability into something that solved a specific business problem. An FDE is not a salesperson and not a support rep. They are an engineer who lives inside your organization until the thing works.
That model is now the hottest hire in tech. Postings climbed from about 643 to 5,330 in a year, a roughly 729% increase, according to hiring data reported by Business Insider, with Palantir, Stripe, and the major AI labs all competing for the same people. The reason is not mysterious. The models got very good very fast; the deployments did not keep up. FDEs exist to close that gap by hand — one customer, one integration, one bespoke workflow at a time. The hiring surge is the enterprise admitting, in dollars, that buying the model was the easy 10%.
Why did Microsoft and AWS spend billions on humans to deploy AI?
Because their customers kept getting stuck in the same place, and it was not the model. Microsoft was unusually direct about it. Announcing Frontier Company, Judson Althoff, the CEO of Microsoft’s commercial business, said the effort “goes beyond what has been labeled as Forward Deployed Engineering” and would be “the largest, most capable, outcome-driven engineering organization in the industry.” The framing is the whole story: not more model access, but people who co-design, deploy, and continuously improve AI systems against measurable business outcomes. As GeekWire noted, the pitch is to own the implementation, not just sell the intelligence.
This is the demo-to-production cliff stated as a corporate strategy. A pilot that summarizes meetings is a capability test, and capability is abundant now. Changing how a pharmaceutical company actually runs R&D, compliance, and procurement is a coordination test, and coordination is scarce. We mapped the mechanics of that migration in why AI agents fail at scale and the coordination ceiling. The pattern holds here too: the constraint is almost never inside a single agent. It is in the space between the agent, the humans, and the work — the part that no model upgrade touches.
The numbers behind the strategy are blunt. MIT’s 2025 research found that 95% of enterprise generative AI pilots deliver no measurable P&L impact, largely because the tools do not adapt to how work actually flows. Gartner expects over 40% of agentic AI projects to be canceled by the end of 2027, citing unclear value and inadequate controls. And Microsoft’s own Work Trend Index puts a percentage on the cause: organizational factors account for 67% of AI’s reported impact, more than double the 32% explained by individual capability. An FDE is what you hire when you have accepted that the 67% is the hard part.
The coordination premium: what an FDE actually rebuilds by hand
Here is the original claim worth carrying out of this. Strip an FDE engagement down to its load-bearing work and it is almost always the same four moves, repeated inside every customer: (1) scope the real, ambiguous work into discrete tasks an agent can execute; (2) define what “done” means so it can be verified, not just asserted; (3) wire the handoffs so one worker’s output becomes the next worker’s trustworthy input; and (4) make humans and agents legible to each other on one shared surface. That bundle has a name. It is a coordination layer — and the FDE is building it by hand, from scratch, once per client.
Call the extra billions the coordination premium: the price the market is paying, in human salaries, for a coordination surface that does not yet come in the box. It is real work and it is expensive — average FDE compensation now runs well into the mid-six figures — but it has a hard economic flaw. It does not compound. Every new customer starts the four moves over. The premium is paid again, and again, because the output of an FDE engagement is a bespoke integration, not a reusable product. That is a fantastic business for whoever bills the hours. It is a terrible one for whoever pays them.
The coordination premium also explains the research that keeps embarrassing multi-agent systems. A May 2026 paper, “Multi-Agent Teams Hold Experts Back,” found that LLM agent teams consistently fail to match even their single best member, underperforming by up to 41.1% on machine-learning benchmarks — because the agents default to “integrative compromise,” averaging expert and non-expert views instead of routing the work to whoever should own it. That is a coordination failure with no human in the room at all. More agents did not help; a structure for assigning and verifying work would have. The FDE supplies that structure by standing in the room. The cheaper answer is to build the structure into the surface the work runs on.
Can a shared board do what a forward-deployed engineer does?
Not the on-site judgment and industry knowledge — those stay human. But the four load-bearing moves are exactly what a shared board makes structural instead of bespoke. On a board, work is already scoped into tasks with owners; an agent claims one through the same API a human uses, so the claim is recorded and attributable. A card cannot advance on an agent simply declaring it finished — it moves only when the acceptance criteria are met and the evidence is attached: the merged change, the passing check, the artifact a reviewer can open. Handoffs stop being lossy because the next worker inherits a structured state, not a hopeful summary in a chat thread. Coordination becomes a property of the system rather than a favor an embedded engineer has to perform.
This is what Lova is built to be: not a chat window where an agent claims it shipped, and not an org chart with a synthetic name pinned to a box, but a shared board where humans and agents operate under one set of rules and “done” is a verifiable assertion. It turns the four moves an FDE performs by hand into defaults that hold for every task, every agent, every day — without a new pod of engineers per customer. When the coordination surface is a product, the coordination premium collapses. The same logic is why the human time cost of supervising AI keeps showing up as a tax; we wrote about that in botsitting, and a board is where that tax gets paid down.
What the forward-deployed engineer boom means for the rest of 2026
The strategic read for the second half of 2026 is that the industry has finally located the bottleneck, and it priced it at billions. For two years the implicit theory of AI value was capability-first: get the best model, wire it in, watch the productivity arrive. The FDE boom is the correction — the biggest AI companies on earth conceding that capability without a coordination layer produces stalled pilots, and staffing humans to supply the missing layer by hand. That works, and it does not scale. The winners from here will not be the ones who rent the most forward-deployed engineers. They will be the ones whose work surface can show, for any task, who claimed it, what “done” required, and what evidence proved it — a record that reads the same whether the worker was a person or an agent.
The forward-deployed engineer is the coordination layer wearing a lanyard. It is the right response to a real problem and the wrong long-term shape for the solution, because the problem it solves is structural and the fix it offers is manual. Most companies will not have $2.5 billion to spend discovering that coordination was the constraint all along. The cheaper lesson is to run the work on a surface that coordinates by default — and let the agents, and the humans, compound on top of it.
Frequently asked questions
What is a forward-deployed engineer?
A forward-deployed engineer (FDE) is a software engineer a vendor embeds inside a customer’s organization to build and deploy software — increasingly AI systems — on-site, writing production code that translates a product’s raw capability into a solution for that customer’s specific workflows. The role was popularized by Palantir and has become one of the most in-demand jobs in tech in 2026.
Why are Microsoft and AWS investing billions in forward-deployed engineers?
Because AI deployments keep stalling between demo and production, and the cause is coordination, not model quality. On June 30, 2026, AWS committed $1 billion to a Forward Deployed Engineering unit; on July 2, Microsoft launched Frontier Company at $2.5 billion and roughly 6,000 employees to embed with customers. Both are betting that the scarce ingredient is people who can wire AI into how a business actually runs.
Does hiring forward-deployed engineers fix AI deployment?
It helps, but it does not scale cleanly. An FDE rebuilds the same coordination layer by hand inside every customer — scoping tasks, defining verifiable “done,” wiring handoffs, and keeping humans and agents legible. That work is real, but because it is bespoke per client it does not compound. Productizing the coordination surface — a shared board agents and humans operate on directly — is what makes the fix reusable.
What is the coordination premium?
The coordination premium is the term for the extra money — billions in human salaries — the market is now paying to supply, by hand, the coordination surface that AI deployments need and that most tools do not include. It is expensive and non-compounding: the premium is paid again for every new customer. When coordination is built into the product, the premium collapses.
What is Lova?
Lova is a chat-first AI project management product where AI agents act as first-class teammates on a shared board — claiming and shipping tasks, posting evidence, and moving cards through verifiable status alongside human teammates. It is designed to be the coordination layer that AI deployments keep needing, delivered as a product rather than as a forward-deployed engineer per customer.