Forward deployed engineers exist because enterprises do not own their own context

Job postings for forward deployed engineers, specialists who embed inside a client's own operations to get AI systems working in production, grew by more than 800% between January and September 2025, according to the Financial Times. That is not a hiring trend. It is a symptom. Enterprises are spending heavily on human beings to do, by hand, something their AI infrastructure should be doing on its own: understanding the business well enough to act on it.

The pattern is familiar: a pilot succeeds, then production stalls.

Most enterprise AI adoption follows the same script. A team scopes a narrow pilot, hand-assembles just enough business context to make it work, and gets a result strong enough to justify wider investment. Then the project moves toward companywide deployment and stalls, because the context that made the pilot work was never captured anywhere durable. It lived in a handful of people's heads for the length of the pilot, not in a governed asset the rest of the organisation could draw on.

Research from MIT's NANDA initiative, drawing on more than 300 publicly disclosed enterprise AI initiatives, 52 structured interviews, and 153 survey responses from senior leaders, found that 95% of organisations running generative AI pilots in 2025 saw no measurable return on the investment. The gap was not model quality, today's frontier models are highly capable. It was that the conditions needed to scale, ownership of the business context the AI depends on, were never established in the first place.

Forward deployed engineers are the current, expensive patch for that gap.

This is the specific gap forward deployed engineers now fill. They learn a client's environment, its terminology, its approval chains, its exceptions, then encode that knowledge into the AI system by hand. It works, narrowly, but it does not transfer. Each engagement is essentially a bespoke integration project, and whatever the engineer learns rarely carries over to the next deployment or the next client. The enterprise ends the engagement no more in control of its own context than when it started, only lighter in budget and temporarily unblocked.

The gap is not closing. It is getting more expensive to ignore.

S&P Global Market Intelligence, surveying more than 1,000 organisations across North America and Europe, found that 42% of companies abandoned most of their AI initiatives in 2025, up from 17% the year before. Abandonment is not slowing as enterprises gain experience with AI, it is accelerating. That is the opposite of what a real skills gap or a maturity curve would produce. It is closer to what you would expect if the underlying problem, context that is never captured as a reusable, governed asset, were structural rather than temporary.

Hiring more forward deployed engineers scales the cost, not the fix.

The obvious response, more forward deployed engineers, more pilots, more bespoke integration work, treats the symptom as though it were the disease. Each additional engagement adds headcount and cost without ever leaving the enterprise with something it owns afterward. As organisations move from single-task AI assistants toward agents that act continuously across multi-step workflows, that gets more dangerous, not less. An agent that lacks grounding in business context does not fail once, it compounds a bad assumption across every step it takes, and the enterprise still has no governed layer to check its work against.

What matters for understanding a codebase is the context that lives inside the ALM cycle itself.

Building and maintaining software well depends on absorbing a specific kind of context: the code itself, its history, the decisions behind it, and everything that happens across planning, building, testing, and operating it, together known as the ALM cycle. Getting an AI system to fully understand a codebase and act on it reliably means focusing on exactly that context, and leaving aside the context that lives elsewhere in the business, HR records, finance systems, internal memos, since none of it shapes how the code behaves. Capturing context at the scale of the whole enterprise, across every department and every system, is a complex undertaking. It relies on capturing and digesting many different categories of context at once, financial, operational, human, technical, each with its own structure and its own sources. Capturing the context of the ALM cycle remains the foundational step in that undertaking, allowing AI systems to act reliably on the code itself while the wider context is built out around it.

Mobioos replaces the manual patch with a layer of context the enterprise actually owns.

Mobioos closes this gap with a permanent layer of context across the ALM cycle. It captures data continuously from across that cycle, the codebase, the tools and systems that run the development lifecycle, and production signals, and maps it onto the codebase at line level through Forge, turning it into the business intent, operational constraints, and domain knowledge an AI system needs, available to any developer, agent, or workflow the moment it is needed rather than being reconstructed by a specialist project by project.

With that context in place, AI systems across the enterprise operate with the same understanding a forward deployed engineer would otherwise spend weeks supplying by hand. Mobioos Fusion runs behind that continuously, pulling in the operational detail that keeps the layer current as the development cycle evolves, so the foundation does not go stale the way a forward deployed engineer's tacit knowledge does the moment the engagement ends.

The organisations closing the AI pilot gap will not be the ones hiring the most forward deployed engineers. They will be the ones that stopped needing to.