Deloitte named a forward-deployed engineering (FDE) practice back in December 2025. At the time, it read as a consulting rebrand with better vocabulary. Then, between May 4 and July 15 of this year, four technology providers committed roughly $9 billion to putting their own engineers inside their customers’ buildings.
Anthropic opened with a $1.5 billion enterprise services venture alongside Blackstone, Hellman & Friedman, and Goldman Sachs, which launched formally on July 15 as Ode with Anthropic. OpenAI followed a week after the May announcement with the Deployment Company, with $4 billion of initial investment, majority-owned and controlled by OpenAI, and backed by 19 firms including TPG, Brookfield, Capgemini, and McKinsey & Company. Amazon Web Services committed $1 billion on June 30 to a forward-deployed engineering organization. Microsoft closed the run two days later with Frontier Company ($2.5 billion and 6,000 people). Google Cloud added a $750 million partner ecosystem commitment in the same window.
Not one of those dollars went into a better AI model. The providers with the world’s most capable models looked at their enterprise pipelines and reached the same conclusion: The next constraint on growth is no longer just model performance but organizational readiness. The bottleneck sits inside organizations whose operating models have not evolved fast enough to convert AI investment into scaled outcomes.
So think of the FDE as a temporary prosthetic for a cognitive operating model that has not caught up, and it works the way training wheels work: genuinely, immediately, and on the condition that somebody eventually takes them off.
What The Training Wheels Actually Hold Up
Palantir pioneered and popularized the FDE as a technical expert embedded with a customer to build complex software and accelerate adoption. That original definition itself has evolved and continues to do so. I see three tech delivery shifts that FDEs are driving in the AI era:
Where the risk sits. Traditional tech implementation ends at go-live with the customer holding the question of whether anything actually improved. Enterprise AI buyers care about measurable business impact. Did customer service productivity increase? Did supply chain efficiency improve? Did revenue cycles accelerate? AWS now describes a methodology of 45 minutes to an idea, 45 hours to a prototype, and 45 days to a customer engagement, measured against business outcomes rather than billable hours. When the provider is paid on the outcome, the provider absorbs integration risk that enterprises have historically eaten alone.
The nature of the work. Consultants used to configure deterministic software against a requirement document. AWS describes its engineers deploying a semantic layer into the customer’s own account, connecting it to enterprise data sources, enriching metadata, and publishing a governed, versioned knowledge graph. Most enterprises carry decades of business logic living in spreadsheets, exception handling, and the heads of people who have been there 15 years or more. Agents cannot read any of it. The FDE’s real product is an encoded answer to how this company decides things, and the best FDEs are translators before they are engineers.
What flows back. Palantir showed that embedded engineers double as a field intelligence function, routing real customer constraints into product roadmaps faster than any customer advisory board. Anthropic’s June partnership with Tata Consultancy Services commits TCS engineering teams to contribute reusable skills and plugins to the Claude Code ecosystem, beginning with claims adjudication and lending advisory.
A Three-Layered FDE Ecosystem Is Emerging
The market is also becoming more specialized. Every provider will have their own take on FDEs.
Platform providers optimize for consumption and stickiness. Salesforce, ServiceNow, Microsoft, AWS, Databricks, and Palantir increasingly use FDE teams to accelerate adoption and embed their platforms more deeply in customers’ operations. The engineering is real, and so are the benefits of integration. But deep integration creates its own gravity. As data, workflows, governance, and institutional knowledge accumulate around a vendor’s platform, the cost of moving rises and the platform becomes harder to displace.
Model providers optimize for effectiveness. OpenAI, Anthropic, and Google are buying, building, and partnering for FDE capacity so that their own engineers can concentrate on agent architecture, evaluation, and reasoning quality rather than large-scale systems integration — narrower scope, sharper edge.
Service integrators optimize for reach. PwC partnered with Palantir to provide FDEs at scale. Accenture launched a Microsoft FDE practice on March 18 and joined ServiceNow’s program at Knowledge in May. Cognizant stood up an EMEA unit this month targeting 5,000 certified engineers. Platform providers can staff hundreds of embedded engineers; only the global system integrators can staff thousands across thousands of environments. Whether they can do that without reverting to project-shaped delivery is the open question for the next 18 months.
The Real Future Of FDEs Is On Your Payroll
Earlier this year, Uber sent 30 of its most AI-proficient engineers to sit inside finance, legal, and HR. For two weeks at a stretch, they watched how the work got done, because as CTO Praveen Neppalli Naga explained it, you cannot automate these processes by looking at process diagrams or documentation. Sixteen of these “agentic pods” ran over two months. Financial pacing reports dropped from two days to 10 minutes and capital allocation across 150 cities from 15 hours to 30 minutes. But Uber did not rent that FDE capability — it built it.
Temporary FDE engagements are the right answer today, when almost no enterprise can staff this itself. But they will not define the next decade. Pivotal Labs ran a similar playbook for product teams more than a decade ago, embedding with clients until it could run the practice “without us.” What the product-centric team was to digital transformation, the forward-deployed engineering team will be to AI: cross-functional, permanent, accountable for an outcome rather than a project, and staffed by people the company employs.
After all, the wheels do not come off because you stopped needing to stay upright. They come off because the balance moved inside. Two years from now, you should count the provider’s engineers in your building, then count your own.
Let’s Connect
I recently published a research report on FDE adoption and will continue researching this area. If your organization evaluates AI delivery models, I would welcome a discussion. The same applies if you consider provider-embedded engineering teams. It also applies if you assess FDEs within your technology services strategy. The model is evolving quickly. Leaders must understand where it creates value and where it does not. That understanding can materially improve AI outcomes. My recent research shows that there is much more to this model than most organizations have realized yet.
