Forward Deployed Engineer: the role that takes AI from pilot to production
A year ago the term only appeared in Palantir job postings. Today OpenAI, Anthropic and Google are hiring Forward Deployed Engineers by the hundred, and banks are following suit. Behind the slightly martial title lies a simple idea: for AI to deliver a result in a company, you need to send engineers to the client, into its data and processes, rather than hand over a platform and a manual. Here is why this model is taking hold, and what it means for an organisation that wants to move from pilot to production.
What a Forward Deployed Engineer is
A Forward Deployed Engineer, or FDE, is a software engineer who works physically and functionally inside the client organisation. They do not sell, and they do not advise in the classic sense: they build, within the client's systems, the solution that makes their employer's product work on a real problem. They feed back to the product team what they learn in the field, and they leave once the solution runs.
The model comes from Palantir, which put it in place from its first contracts with US intelligence agencies. According to the Pragmatic Engineer newsletter, quoted by MarkTechPost in May 2026, Palantir's first embedded engineers were nicknamed "Deltas", and until 2016 the company had more FDEs than classic software engineers. The launch of the AIP platform in April 2023, followed by three-to-five-day "bootcamps" where FDEs build a use case with the client, gave the model a second life: by the end of 2024, more than a thousand bootcamps had been run, with a high conversion rate into contracts. In the first quarter of 2026, Palantir reported 85% revenue growth and 133% growth in its US commercial business.
Why the AI labs are copying the model
The reason fits in one number. According to the MIT NANDA report "State of AI in Business 2025", 95% of enterprise generative AI pilots show no measurable impact on results. The models are powerful, the demos impressive, and yet the value does not arrive. What is missing is not the model: it is integration with the organisation's data, systems, processes and habits.
The labs drew the conclusions. OpenAI built its FDE team in late 2024 and accelerated hiring in 2025, then announced on 11 May 2026 "The Deployment Company", a joint venture that raised more than 4 billion dollars from nineteen investors, with partners such as Goldman Sachs, BBVA, McKinsey and Capgemini, and the acquisition of the consultancy Tomoro and its roughly 150 engineers. The BBVA case is telling: starting from a ChatGPT Enterprise rollout, the bank now serves 120,000 employees in 25 countries with the support of OpenAI FDEs.
A week earlier, on 4 May 2026, Anthropic announced with Blackstone, Hellman & Friedman and Goldman Sachs an AI services company valued at 1.5 billion dollars. The words of its CFO, Krishna Rao, sum up the situation: "Enterprise demand for Claude is significantly outpacing any single delivery model." Marc Nachmann of Goldman Sachs spoke of "democratizing access to forward-deployed engineers" for mid-sized companies.
The labour market is following. A compensation report published on 21 May 2026 covering 1,200 FDEs gives a median total compensation of 385,000 dollars at mid level in the labs, and notes that New York concentrates 35% of postings, driven by finance and regulated sectors. FDEs now earn 10 to 25% more than machine learning engineers at senior levels. That says something: scarcity is no longer in modelling, it is in deployment.
What the model changes for a bank, an insurer or an SME
The FDE is not a Silicon Valley fad, it is the recognition of a fact integrators have long known: value is created at the junction between technology and business, and that junction cannot be outsourced remotely.
For a client organisation, three practical consequences.
The pilot must be designed as production. An FDE does not run a demo, they connect the model to real data, under real security constraints, and measure a business result. That is exactly what the 95% of pilots without impact were missing.
Knowledge must stay in the company. The risk of the FDE model is dependency: the engineer leaves with the knowledge. The organisations that benefit most plan from day one for an internal counterpart, living documentation and a gradual handover of maintenance.
Governance is not negotiable. Forbes noted in April 2026 that FDEs excel in dynamic systems where they can adjust in real time with business teams, but that in stable systems with strict governance the approach "may introduce risks and is not always appropriate". In a bank or an insurer, the FDE must work within the model risk management framework, not beside it.
A concrete situation: a financial institution that tested a document-processing assistant for six months without connecting it to its case management system. By embedding two engineers in the business team for ten weeks, with access to real data and a quantified target on processing time, the assistant goes into production and the lead time is halved. The technology did not change, the way it was delivered did.
Consulting becomes embedded engineering
This shift is also redrawing the consulting profession. Firms that deliver recommendations and leave are losing ground to those that deliver running systems. At 39 Advisory, this has been our model from the start: our FNZ Studio, Java and data consultants work inside our clients' teams, on their platforms, with delivery targets. The FDE wave does not surprise us, it gives a name to what we do, and it confirms that the need is no longer limited to large groups.
What we take away
The Forward Deployed Engineer is the market's answer to the failure of AI pilots: the technology was ready, the delivery was not. For a company, the lesson is simple. Stop buying a platform and a manual, buy a result, delivered by engineers who work in your systems, within your governance, and who hand over their knowledge before they leave.
If you have a pilot that never reaches production, let's talk: it is usually a question of delivery model, not of AI model.
Comments