The 2025 enterprise AI conversation was about the model. The 2026 conversation is about the deployment. The signal is converging from three independent directions, and it points to the same conclusion: the constraint on enterprise AI value is no longer the capability of the underlying model — it is the engineering work required to wire that capability into real systems, with real data, real users and real failure modes.
The most public marker came on **11 May 2026, when OpenAI announced the OpenAI Deployment Company — branded DeployCo — as an autonomous business unit with more than four billion dollars of initial investment**. The structure was reported across the technology press, including **SiliconANGLE** and **The Register**. Per those reports, **19 investors backed the launch, including Bain & Company, McKinsey & Company and Capgemini** — three of the world's largest consulting and integration firms taking equity in the same vehicle. DeployCo's stated remit is to embed **Forward-Deployed Engineers (FDEs)** directly inside customer organizations to build production AI systems on top of OpenAI's models, in customer stacks, on customer data.
That OpenAI is now selling the **engineers**, not only the API, is a statement about where the value is. So is the participation of the three integrators on the cap table. So is the operational evidence from the field: per the **LangChain State of Agent Engineering report**, **roughly 57% of organizations now have AI agents in production**, and **the most commonly cited blocker — at 32% — is quality and integration**, ahead of model capability, ahead of cost, ahead of regulation.
The story this set of data points tells is simple. The model is not the bottleneck. The bottleneck is the deployment. And the new operating model — large vendor and large consultancy alike — is to put senior engineers **inside the customer**, for as long as it takes, until the system works in production.
This article is a builder's view of that shift. What forward-deployed engineering actually is, why it has emerged as the dominant pattern in 2026, what it costs at the DeployCo / McKinsey tier, and how the mid-market — which cannot afford that tier — accesses the same model from a nearshore hub.
The term **Forward-Deployed Engineer** entered enterprise vocabulary through Palantir in the 2010s and was reused by a handful of AI-native companies through 2024 and 2025. DeployCo's 2026 launch crystallized it. The role description across the firms using it has converged on a recognizable shape:
A senior engineer — typically 5+ years of production experience, with a deliberate generalist profile across distributed systems, data engineering, ML operations and product engineering — who is **embedded in a single customer for an extended engagement** (typically six to twelve months, often renewed). The FDE does not work from a vendor office on a vendor roadmap; the FDE sits in the customer's Slack, the customer's Jira, the customer's GitHub, and ships against the customer's backlog. The FDE is paid by the vendor or the consultancy, but accountable to the customer's outcomes — model in production, integration working, users adopting, metrics moving.
The pattern is a deliberate departure from two prior models. It is not a **statement-of-work consultancy** — there is no deliverable list signed at the start and re-litigated at every change. It is not a **staff-augmentation contractor** — the FDE is not a body on a billing rate, the FDE is a senior engineer accountable for an outcome. It is closer in shape to the **embedded technical co-founder** model the venture world is familiar with, applied to enterprise AI deployment.
The reason it has emerged now is structural. The 2024–2025 wave of enterprise AI projects ran into a consistent failure mode: the model worked in a notebook, the pilot worked in a demo, and the production system was never built. The integration work — connecting the model to the data, the data to the users, the users to the workflow, the workflow to the metrics, the metrics to the governance — turned out to be **80% of the timeline and 90% of the failure rate**. A model-only vendor selling APIs cannot deliver that 80%. A consultancy selling a deck cannot deliver that 90%. An embedded engineer can.
Three data points, taken together, explain why DeployCo, McKinsey, Bain, Capgemini and the rest are all converging on the embedded-engineer pattern at the same time.
**The 57% adoption number.** Per LangChain, roughly 57% of organizations have agents in production. That is a step change from the 2024 baseline. It also means that the population of enterprises with **real production AI workloads to operate** is now the majority, not the early-adopter minority. Each of those workloads needs maintenance, not just a launch.
**The 32% integration-and-quality bottleneck.** The same report identifies integration and quality as the single largest blocker, ahead of model capability. That is the FDE's specific addressable problem. It is not a problem the vendor can fix in the API.
**The four-billion-dollar DeployCo round, with three integrators on the cap table.** OpenAI is the largest model vendor in the world. Bain, McKinsey and Capgemini are three of the largest integrators in the world. Their joint signal that the deployment layer is worth four billion dollars of investment, structured as an embedded-engineering vehicle, is the clearest possible statement that the gap is real and durable.
The combination forces a new operating model. The 2024 model — buy an API, hire a few contractors, hope it ships — does not survive contact with the 32% blocker. The 2026 model puts senior engineers inside the customer until the system works.
The DeployCo and consulting-major version of forward-deployed engineering is expensive. Public-domain reporting on consulting day rates for senior AI engineering deployments in 2025 and 2026 puts the typical loaded day rate for an FDE-equivalent profile at major US and Western-European integrators in the **$2,500 to $4,500 per day** range, and frequently higher for partner-led engagements. A single forward-deployed engineer at that rate, on a twelve-month engagement, lands between roughly **$600,000 and $1.1 million** of consulting fees before any platform or model spend.
For a Fortune 500 organization with a strategic AI program and a CFO who has already signed the platform invoice, that is a defensible number. For the **mid-market** — companies in the €50M to €500M revenue range that account for the bulk of the 57% adoption figure — it is not. The mid-market needs the same operating model at a fraction of the cost, or it does not get the operating model at all.
This is where the geography of the deployment layer matters. **The forward-deployed engineering model is portable.** What makes it work is the seniority of the engineer, the duration of the embedding, the access to the customer stack and the accountability to the customer outcome. None of those depend on the engineer being in Manhattan or Munich.
The economics of running the same model from a nearshore hub with European time-zone overlap change the cost structure by an order of magnitude. A senior engineer in Morocco — five-plus years of production experience, native English and French (often Spanish or Arabic), Central European Time year-round, two hours by flight from Frankfurt — bills at **roughly fifteen to twenty-five euros per hour** in a sustained engagement. A full-time embedding lands between **€30,000 and €50,000 per engineer per year** of loaded cost to the client, depending on seniority and contract structure. That is one-tenth to one-fifteenth of the equivalent DeployCo / consulting-major rate.
The talent-pool depth supports the model. Morocco's positioning at **rank 26 worldwide on the Ataraxis Global Outsourcing Talent Index 2026** — per **Morocco World News** and the **Atlas Brief** reporting on the index — places it in the top tier of global outsourcing talent destinations, with a comparable cost basis to India and the Philippines and a substantially better time-zone alignment for European customers. The full positioning of our nearshore delivery is in [why Morocco](/en/why-morocco).
The mistake the mid-market makes when it tries to access the forward-deployed model on a nearshore budget is to default to a generic staff-augmentation contract — a body shop billing rate, a daily timesheet, no outcome accountability. The result is the same failure mode the FDE model was invented to solve.
The structure that has worked for our clients through 2026 is closer to the DeployCo shape than to a staff-augmentation contract. The components:
**Senior profile, not junior.** The embedded engineer is a generalist with production experience across data, ML operations and product engineering. The headline rate per hour is not the variable — output per week is. A junior engineer at half the rate ships a quarter of the value; the math does not work.
**Single-customer embedding.** The engineer is dedicated to one customer for the engagement. No multi-account splitting, no "we will get to your ticket next week." The Slack is the customer's Slack, the standup is the customer's standup.
**Outcome-aligned objectives.** The engagement is scoped against the customer's production metrics — agent in production, integration shipped, deflection rate moved, accuracy gate met. The output is the metric, not the timesheet.
**Time-zone overlap, not asynchronous handoff.** A forward-deployed engineer who answers a Slack thread eight hours later is, in practice, not embedded. Central European Time year-round, with intentional overlap into UK and US morning hours, is what makes the model behave like an internal hire instead of an offshore contract.
**A small pod, not a single engineer.** For most production AI deployments, the right unit is two to three engineers — typically a lead engineer with data and ML-operations depth, a product engineer for the integration and UI surface, and a part-time architect for the cross-cutting decisions. This is the structure our [dedicated development teams](/en/services/software-development/dedicated-development-teams) service is built around.
**Continuity, not rotation.** The same engineers stay through the engagement. The FDE model breaks the moment the embedded engineer is replaced; institutional context is the asset.
The FDE engagement is most useful at the moment where the customer has chosen a model, has data that matters, and needs to wire it into production. That work spans three engineering tracks that have to ship together.
**The model integration track** — getting the model into the customer stack with the right SDK, the right authentication, the right routing, the right cost controls. Our [AI and machine learning development service](/en/services/software-development/ai-ml-development) is structured around this track and the model selection that precedes it.
**The system integration track** — connecting the model to the systems of record, the workflow tools, the identity layer, the data pipeline, the monitoring stack. This is most of the FDE's day and the place where unprepared engagements stall. The general [custom software development capability](/en/services/software-development) is the broader umbrella for this work.
**The product engineering track** — surfacing the AI capability in the user's workflow in a way that gets adopted, with the guardrails, fallbacks and human-in-the-loop paths that production demands.
A nearshore FDE pod of two to three engineers covers all three tracks in parallel. A single engineer cannot. A consulting deck cannot. A statement-of-work vendor reluctantly takes two months to scope each change. The pod model is what the 2026 enterprise AI workload actually needs.
A nearshore FDE model is not a strict superset of the DeployCo engagement. There are trade-offs a buyer should understand before signing.
**Brand premium.** A McKinsey or Bain engagement carries a board-level brand the nearshore pod does not. For a small number of strategic deployments where the board needs the brand, the premium can be justified.
**Frontier-model proximity.** DeployCo engineers work physically close to the OpenAI model roadmap. For a customer building on the absolute frontier of an unreleased capability, that proximity matters. For the 95%+ of enterprise deployments built on generally-available models, it does not.
**Onshore regulated workloads.** Some specific regulated industries — defense, classified government, certain healthcare segments — require onshore residency for the engineers themselves, not just the data. For those workloads, the nearshore model is not the right answer.
For the broad mid-market enterprise AI deployment, none of these trade-offs apply. The nearshore FDE pod gives the same operating model at a price the mid-market can sustain.
The deployment-side risk this article focuses on — getting the AI into production — is one of two risks the 2026 enterprise has to govern simultaneously. The other is the **security perimeter of the AI tools themselves** once deployed: the copilot you rolled out is a new attack surface, and the vendor's patch cycle does not absolve you of governing access, data and monitoring on your own deployment. The companion piece is [Your AI Copilot Is a New Attack Surface: Securing Enterprise AI Tools in 2026](/en/blog/m365-copilot-flaw-securing-enterprise-ai-tools-2026). Read it alongside this one if your scope covers both build and operate.
For a mid-market enterprise considering the forward-deployed engineering model on a nearshore budget, the path that has worked for our clients is short.
**Days 1–10.** Define the single production AI workload the engagement targets. Document the systems it must integrate with, the data it must access, the metric it must move. Identify the internal owner.
**Days 11–20.** Stand up the pod — typically a lead engineer, a product engineer and a part-time architect. Land them in the customer's Slack, Jira and GitHub. First sprint plan jointly.
**Days 21–40.** Build the integration, the data plumbing and the first version of the user-facing surface. Ship to a controlled user group. Measure.
**Days 41–60.** Iterate against the metric. Add guardrails, fallbacks and monitoring. Hand the system off to the customer's operations team with documentation, runbooks and a defined ongoing engagement.
A pod that runs this sixty-day path lands the system in production, and the customer ends the engagement with a working AI workload, owned internally, supported by a small embedded team at a cost the business can absorb.
The 2026 evidence — DeployCo's four-billion-dollar launch with three integrators on the cap table, 57% of organizations in production, 32% citing integration as the primary bottleneck, OpenAI selling engineers and not only APIs — is converging on a single conclusion. **The deployment layer is where enterprise AI value is created or lost.** The DeployCo / consulting-major answer is forward-deployed engineering at $2,500–$4,500 per day, accessible to a small number of large customers. The nearshore answer is the same operating model — embedded senior engineers, single-customer engagement, outcome accountability, time-zone overlap — at one-tenth the cost, accessible to the mid-market that is actually doing the deployment work. Pick the model. The body-shop contract and the deck-driven engagement are not in the running anymore. ${CTA_FDE_NEARSHORE}
OpenAI announced the OpenAI Deployment Company (DeployCo) on 11 May 2026 as an autonomous business unit with more than four billion dollars of initial investment. Per reporting from SiliconANGLE and The Register, 19 investors backed the launch, including Bain & Company, McKinsey & Company and Capgemini. DeployCo's stated remit is to embed Forward-Deployed Engineers (FDEs) directly inside customer organizations to build production AI systems on customer stacks and customer data.
A Forward-Deployed Engineer is a senior generalist engineer — typically 5+ years of production experience across distributed systems, data engineering, ML operations and product engineering — embedded inside a single customer for an extended engagement (often six to twelve months). The FDE sits in the customer's Slack, Jira and GitHub, is paid by the vendor or consultancy, and is accountable to the customer's production outcomes rather than a fixed deliverable list.
The LangChain State of Agent Engineering report finds that roughly 57% of organizations have AI agents in production and that the most commonly cited blocker, at 32%, is quality and integration — ahead of model capability, ahead of cost, ahead of regulation. The pattern explains why the 2026 operating model has shifted from API-plus-contractors to embedded forward-deployed engineering.
Public-domain reporting on consulting day rates for senior AI engineering deployments in 2025–2026 puts the typical loaded day rate for an FDE-equivalent profile at major US and Western-European integrators in the $2,500 to $4,500 per day range, frequently higher for partner-led engagements. A single forward-deployed engineer on a twelve-month engagement lands between roughly $600,000 and $1.1 million of consulting fees before any platform or model spend.
A senior engineer in Morocco — five-plus years of production experience, native English and French (often Spanish or Arabic), Central European Time year-round, two hours by flight from Frankfurt — bills at roughly fifteen to twenty-five euros per hour in a sustained engagement. A full-time embedding lands between €30,000 and €50,000 per engineer per year of loaded cost. Morocco ranks 26th worldwide on the Ataraxis Global Outsourcing Talent Index 2026 per Morocco World News and Atlas Brief reporting, with a comparable cost basis to India and the Philippines.
Three trade-offs: when a board-level brand premium (McKinsey, Bain) is part of the strategic requirement; when the customer needs physical proximity to an unreleased frontier-model roadmap; or when a specific regulated workload (defense, classified government, certain healthcare segments) requires onshore engineer residency, not just onshore data residency. For the broad mid-market enterprise AI deployment, none of these trade-offs apply and the nearshore pod delivers the same operating model at one-tenth to one-fifteenth the cost.
CALL IT DEV — Software, AI and dedicated tech teams — Casablanca | Madrid | Dubai — contact@callitdev.com — +212-537-373777