Forward-deployed engineer (FDE) job postings on Indeed grew 729% in a single year, from 643 in April, 2025 to 5,330 in April 2026, according to data reported by Business Insider. Anthropic, OpenAI, and AWS have each announced major FDE initiatives in 2026 alone, with AWS committing $1 billion to embed thousands of engineers directly with customers.

Interest from the supply chain community is accelerating. The model promises something chief supply chain officers have wanted for years: enterprise-grade agentic artificial intelligence, deployed fast, built around the specific complexity of their operations rather than configured to approximate it. Tech and service providers are ready to sell this delivery model.

The question a CSCO needs to answer is whether their organizations are ready to buy it.

What Is an FDE?

The role of FDE places engineers directly inside a client organization to build production AI systems against real data, workflows, and operational constraints. They scope the problem, write the code, deploy the solution and stay until it works.

What makes the current wave different is what the engineers are building. FDEs today are constructing agentic AI systems: agents that sense conditions, determine the appropriate action, execute and adapt when circumstances change. The decision logic is embedded in the agents themselves, not encoded in predefined rules that require a specialist to update. That distinction matters.

CSCOs are caught between competing investment pressures to simultaneously maintain aging enterprise software, evaluate new AI-enabled planning platforms, and build internal agent capability with talent that is scarce and expensive. Each of these approaches comes with its own pros and cons.

An FDE offers a fourth option: enterprise-grade agentic AI, delivered by engineers who understand both the technology and your supply chain, built to your specific workflows. The speed-to-value advantage is real. So is the bespoke quality. For organizations without internal AI engineering capacity (which describes most supply chain functions today), FDE may be the only realistic path to deploying agentic AI at meaningful scale.  

The model is also highly attractive to the providers themselves. Whether large tech companies or established supply chain vendors, FDE engagements allow them to demonstrate value fast while securing customer commitment, even when their agentic capabilities are still maturing. For the provider, embedding an engineer delivers something beyond the immediate deliverable: an intimate relationship with your staff, deep account stickiness, and the opportunity to learn from you and your live operational data. 

The Familiar Risks

Elements of the FDE model have precedent. System integrators built software applications, external software engineers customized them and tech providers configured their own off-the-shelf tools. FDE is the next iteration of these relationships between an enterprise and external technology provider. 

CSCOs who lived through those prior waves will recognize some of the risks that FDE introduces. There’s the scope creep driving up initial cost estimates and chosen individuals transitioning to other engagements. There’s the unwieldy portfolio of custom features that require active governance to track, evaluate and retire. And there’s a dependency on the external provider that extends beyond initial expectation.

These are familiar risks, and CSCOs have navigated them in the past with clear contracts, engagement plans and explicit governance structures. What is less familiar is harder to contract around.

The Specific Risk

The FDE model carries additional risks because of the combined nature of the agentic AI technology and the delivery model. They center on a single dynamic: your organization's growing distance from its own supply chain decision logic and rules.

They have always lived somewhere. It started with the planner, analyst, or buyer. They’re the ones who defined how decisions should be made, based partly on their judgment and partly on analytical insights. That logic was then translated into rules inside supply chain applications: reorder points, allocation thresholds, supplier prioritization and approved transportation modes. The rules were visible and auditable by anyone with access to the system.

Agentic AI collapses that layer. The agent senses conditions and determines the appropriate action without predefined rules. The decision logic isn’t encoded in a place your organization controls because it emerges from the model, informed by data and current supply chain goals. Of course, broader ethical and legal guardrails exist. But supply chain performance decisions are increasingly made by agents acting on conditions, not on rules your team defined. Business logic that once lived in your people and your systems is migrating into the agent.

This is true of any agentic AI solution. But the FDE model makes it more acute.

An off-the-shelf agentic solution builds on shared workflows. The logic is documented, updated across the vendor's customer base, and available to the market. You can compare notes with your peers or rely on third-party validation. Your team can benchmark and learn from a broader ecosystem.

FDE builds agents around your specific workflows. That bespoke quality is both an attraction and a risk. The logic and agents are unique to your organization, built by capable FDEs that understand your business. But this means that cohort of peers leveraging the same solutions could be non-existent. It also means that the engineers building those unique agentic AI solutions aren’t your employees and won’t stay indefinitely. The planner and analyst who once defined the logic are now two layers removed from it. First, the logic migrated into the agent; and second, the agent was built by an external provider.

What to Ask Before You Sign

CSCOs are well past debating whether AI matters. The conversation has shifted to investment decisions: given an AI-driven business environment, what are the best bets for generating fast, scalable and sustainable returns? The FDE delivery model is one such bet. It can deliver significant payoffs with the right planning.

To start with, cover the bases. Mitigate the risks the FDE model shares with prior waves of external engineering: build governance structures before the engagement begins, agree on pricing terms that account for the volatility of the AI tech market, and define clear transition protocols for when the engagement ends.

Then, ask the three questions that are specific to what FDE and agentic AI introduce together:

First: When the FDE engagement ends, your organization inherits an agent environment that has no rulebook. Maintaining, modifying and extending it requires machine learning engineering skills that most supply chain organizations don’t have and can’t easily hire. Who owns this environment the day after the engagement closes

Second: Agent performance isn’t static. An FDE-built agent that works well at go-live will drift as conditions change, data shifts and exception cases accumulate. Unlike traditional software, where a broken rule can be found and fixed, diagnosing why an agent is underperforming requires interpreting model behavior over time. Does your organization have a plan for monitoring performance and auditing decisions when there’s no vendor support desk to call?  

Third: Your planners will spend weeks sitting alongside FDE engineers, translating decades of operational judgment into agent logic. That process moves knowledge in one direction. What is the explicit plan to keep it from walking out the door when the engagement ends?

FDE interest reflects a real gap between the agentic AI capability that supply chains need and the internal engineering talent most of them have. For many organizations, it will be the fastest credible path forward.

But fast isn’t the same as ready. The CSCO who walks into an FDE engagement knowing what they’re gaining, what they’re giving up and who owns the machine when the engineers leave is in a fundamentally different position than the one who signs on the strength of the demo. The questions above aren’t justification for walking away; they’re the conditions for making it work.

Noha Tohamy is an independent supply chain AI adviser and creator of the Decision-First Operating Model.