The Deployment Gap

Two things are true at once in the AI industry right now, and they're in tension with each other.

The first: AI products are not plug and play. Across the 89 AI companies in our dataset, more than 1,300 job postings — roughly 16% of all hiring — exist in the space between "product ships" and "customer uses it in production." That's more than half the size of the entire engineering function at these companies. The products work, but getting them to work inside a specific customer's environment requires an enormous amount of technical, human effort.

The second: the role that's come to symbolise this problem — the Forward Deployed Engineer — has become one of the most talked-about jobs in tech while remaining one of the least understood. The title means very different things at different companies, and it's only one piece of a much larger organisational pattern that most coverage has missed.

This article maps out exactly what that pattern looks like, grounded in 8,648 job postings across 89 AI companies.

---

The Forward Deployed Moment

Monthly job listings for Forward Deployed Engineers increased by more than 800% between January and September 2025, according to an analysis by Indeed and the Financial Times. Andreessen Horowitz called it "the hottest job in startups" in a June 2025 piece titled "Trading Margin for Moat," arguing that enterprises buying AI need hands-on support to operationalise models into real-world solutions — and that companies providing this support, even at the cost of lower margins, build more durable competitive positions.

Palantir pioneered the role in the early 2010s. Until 2016, the company employed more FDEs than software engineers. What was once a single company's idiosyncratic staffing choice has become a category: in our dataset, 147 active FDE postings span 31 companies, from Palantir's 41 to Anthropic's 4 to single postings at smaller firms.

Compensation data from Pave reinforces the role's technical foundation. FDEs earn 40.3% more than professionals in traditional Customer Success and Implementation roles at equivalent job levels, and just 5.3% less than software engineers on a level-normalised basis. The market is treating this as an engineering job with customer fluency, not the reverse.

But here's the problem with the FDE narrative: it treats "deployment" as one thing. It isn't. And the FDE — despite the attention — represents only 147 of the 1,360 deployment-spectrum jobs in our data. The other 1,200 are doing related but distinct work that rarely gets discussed with the same specificity.

---

The Full Deployment Spectrum

Ten roles make up the deployment spectrum at AI companies. They span four different organisational functions — Sales & GTM, Customer Support, Engineering, and Product. The fact that deployment work is distributed across the org chart, not concentrated in one team, is itself a finding.

The largest single role in this spectrum isn't the Forward Deployed Engineer — it's the Solutions Architect, with 426 jobs across 43 companies. Solutions Engineers add another 221. Together, SA and SE account for nearly half the entire deployment workforce. These are the roles that design how an AI platform fits a customer's existing systems, validate feasibility, and build proof-of-concept implementations before the deal is signed. The FDE, by contrast, typically arrives after the deal closes, writing production code inside the customer's environment.

Why does this matter? Because the public conversation about "AI deployment roles" has collapsed around a single job title, when the reality is ten distinct roles with different skill profiles, different organisational homes, and different points in the customer relationship.

---

What a Company Sells Determines Everything

The 89 companies in this dataset are not a homogeneous group. They include pure research labs, GPU infrastructure providers, data platforms, vertical applications, and companies that span multiple categories. What a company sells — and how much tailoring its product requires — shapes which deployment roles it hires and how many.

Pure research labs have almost no deployment roles. They sell models or APIs — the integration burden falls on the customer. Infrastructure providers lean on Solutions Architects because they sell complex platform products that require architecture design, not hands-on code deployment. Application companies have the broadest deployment teams: CSMs, Implementation Specialists, Engagement Managers, and most of the Forward Deployed PMs. And the Palantir cluster — Infrastructure + Application — is FDE-dominant, reflecting a model built around embedding engineers directly.

The deployment gap is real across the board, but its shape differs depending on what a company sells.

---

The Customer Lifecycle

These roles map to five phases of the customer relationship.

Discover and Evaluate. Client Partners identify whether a company's AI product is a plausible fit for a prospective customer's environment, then hand off to Solutions Architects and Solutions Engineers who design the technical approach and validate feasibility. At this stage, the work looks like architecture diagrams, proof-of-concept builds, and technical discovery calls. Together, these three roles account for 706 active jobs — more than the entire deployment-phase workforce.

Close. In AI sales, the deal often hinges on a technical proof point rather than a pitch deck. Solutions Engineers and Solutions Architects frequently build working demonstrations in production-adjacent environments. This is where the 647 combined jobs across SA and SE do much of their work: making AI tangible enough that a procurement team will sign.

Deploy. Four roles converge here, and this is where the boundaries get confused. Forward Deployed Engineers write production code inside customer environments. Implementation Specialists configure and integrate products into existing workflows. Engagement Managers coordinate multi-workstream deployments across customer teams. Forward Deployed PMs shape AI agent behaviour for specific customer contexts. Together, these roles represent 261 jobs across 68 companies, and the boundaries between them are drawn differently at every organisation.

Scale. Customer Success Managers drive adoption, retention, and expansion after deployment. With 150 active postings across 30 companies, this is the most recognisable of the deployment roles, though at AI companies the technical bar is often higher than in traditional SaaS. OpenAI, for instance, titles the role "AI Success Engineer" — a title that signals engineering fluency alongside account management.

---

Why the Boundaries Blur

We categorise every job in our dataset into a standardised role taxonomy — grouping the hundreds of unique titles companies use into a consistent set of roles based on what the job description actually asks for. That process reveals patterns that would be invisible if you took titles at face value.

Same title, different job. OpenAI posts dozens of jobs with the title "AI Deployment Engineer." Based on the job descriptions, these map to three different roles: Solutions Engineer, Solutions Architect, and Implementation Specialist. One variant — "AI Deployment Engineer, Codex" — describes what is essentially a Solutions Engineer in one posting and a Solutions Architect in another. The descriptions genuinely differ. When the emphasis is on customer-facing API integration, the work is SE work. When the emphasis shifts toward system architecture and media partnerships, it's SA work.

Similarly, "Forward Deployed Engineer" postings from RunPod and OpenAI describe work that looks more like Solutions Engineering — pre-sales validation, integration scoping — than the embedded production engineering that defines the FDE role at Palantir or Anthropic. Palantir's own "Forward Deployed Engineer — Strategist" variant reads more like a Solutions Architect role, reflecting a shift toward advisory work.

Invented titles. At least six companies in the dataset have created proprietary titles for roles that map to existing categories. Harvey uses "Legal Engineer" for a role that, based on the description, functions as a Solutions Architect — designing how AI fits into a law firm's document review workflow. Cognition calls theirs "Delta Engineer" — a role whose description reads like a Solutions Engineer focused on pre-sales technical validation. Sierra titles it "Product Manager, Agent Development" for what functions as a Forward Deployed PM. ElevenLabs and Mistral use "Deployment Strategist." OpenAI alone uses "AI Deployment Engineer," "Demo Experience Engineer," and "AI Success Engineer" across three different lifecycle phases.

This isn't sloppiness. It reflects companies trying to name work that doesn't fit existing categories.

The SA/SE overlap specifically. 647 combined jobs across these two roles, and the boundaries are porous. Jobs titled "Solutions Engineer" sometimes describe architecture-level work. Jobs titled "Solutions Architect" sometimes describe hands-on demo building. Both roles conduct technical discovery and build proof-of-concept implementations. The difference is emphasis — architecture design versus hands-on demonstration — not a clean boundary. At Databricks, the distinction is clearer (171 SAs, 45 SEs, with Resident Solutions Architects working in specific industry verticals). At smaller companies, a single person often does both.

---

What the Tech Stacks Reveal

When titles are unreliable, the technology requirements are a more useful guide to what a role actually involves.

The distinction between Forward Deployed Engineer and Solutions Architect is sharpest at the technology layer. FDE postings require PyTorch, RAG, LangChain, multi-agent systems, and fine-tuning — hands-on ML engineering tools. SA postings require Databricks, Apache Spark, Delta Lake, MLflow, and data warehousing — platform architecture tools. Both require Python and SQL, but the surrounding stacks diverge.

Solutions Engineers sit between the two. Their technology profile emphasises API integration, LLMs, and generative AI, with cloud infrastructure and REST APIs. They need enough model understanding to demonstrate capabilities and enough systems understanding to validate integration feasibility, but they're not deploying models into production (FDE) or designing data architectures (SA).

A job requiring PyTorch and RAG is FDE work regardless of what the company calls it. A job requiring Databricks and Spark is SA work. For anyone evaluating roles in this space, the technology requirements in the posting are a more reliable signal than the title.

---

How Different Companies Staff This Spectrum

What a company sells determines which deployment roles it hires.

Selling a complex data/AI platform (the Databricks model). SA-dominant. Databricks has 171 Solutions Architects and only 12 Forward Deployed Engineers — a 14:1 ratio that reflects work centred on architecture guidance rather than hands-on code deployment. Customers need help designing how the platform fits their data estate. The SA role here is advisory, not embedded.

Selling a product with customer-specific AI configuration (the Harvey, Glean, Notion model). Balanced teams across CSM, SE, Implementation, and often FDPM. Application companies collectively employ 73 CSMs, 31 Implementation Specialists, and 28 Engagement Managers — the highest numbers in those roles of any company type. The product works broadly, but AI features require domain-specific setup. Glean illustrates this: 14 CSMs, 9 Engagement Managers, 5 Solutions Engineers, 3 Solutions Architects, 3 Forward Deployed Engineers, and 3 Forward Deployed PMs — a deliberate spread across the entire lifecycle.

Selling an FDE-model platform (the Palantir model). FDE-dominant. Palantir has 41 Forward Deployed Engineers — more than any other single role. Their product requires engineers embedded directly inside customer operations. The deployment model is the product strategy.

Selling an agent-first product (the Sierra model). FDPM-dominant. Sierra has 7 Forward Deployed PMs and 6 Solutions Engineers, but no Solutions Architects and no Forward Deployed Engineers. Their deployment model centres on shaping agent behaviour per customer — product work rather than engineering work.

Building the deployment function in real time (the OpenAI model). OpenAI has 100 deployment-spectrum roles spread across FDE (30), SE (27), Client Partner (17), SA (9), CSM (8), and Implementation Specialist (7). No single role dominates. Compare that to Databricks's SA-heavy structure or Palantir's FDE-heavy structure, and the pattern is clear: OpenAI hasn't yet converged on a single deployment model. They're experimenting across all of them simultaneously, inventing three different naming conventions for three different lifecycle phases.

The deployment gap exists across nearly all company types, but the shape differs. An application company's deployment challenge is customer onboarding and adoption. A platform company's is architecture design. An FDE-model company's is hands-on integration in messy environments. A reader evaluating roles in this space should consider not just which role they want, but what type of company they'd be doing it at — because a Solutions Architect at Databricks and a Solutions Architect at a 20-person AI startup are different jobs.

---

Choosing a Path

For anyone evaluating roles in this space, six paths present themselves.

Forward Deployed Engineer — for engineers who want to write production code in customer environments, working at the model and agent layer. See: Forward Deployed Engineer, Applied AI — Anthropic, Forward Deployed AI Engineer — Palantir, Forward Deployed Engineer, Agentic Platform — Cohere.

Solutions Architect — for those who design how AI platforms fit into enterprise data and infrastructure. See: Solutions Architect, Applied AI — Anthropic, Resident Solutions Architect, Financial Services — Databricks.

Solutions Engineer — for those who bridge technical demonstration and integration feasibility, working at the API and LLM layer. See: AI Deployment Engineer, Codex — OpenAI, AI Deployment Engineer, Codex — OpenAI (same title, different posting — one describes SE work, the other SA work, illustrating the boundary in practice).

Engagement Manager / Implementation Specialist — for those who coordinate and execute multi-workstream deployments. See: Engagement Manager (NYC) — LangChain, Engagement Manager, AI Solutions — Snorkel AI, Engagement Manager — Databricks.

Forward Deployed PM / AI PM — for product managers who shape AI agent behaviour in customer-specific contexts. See: Product Manager, Agent Development — Sierra, Product Manager, API Agents — OpenAI, Staff Product Manager, Agents — Harvey.

Customer Success Manager — for those who drive adoption, retention, and expansion after deployment. See: Customer Success Manager, Industries — Anthropic, Enterprise Customer Success Manager, EMEA — Harvey, AI Success Engineer — OpenAI.

---

Observations

The deployment apparatus is more than half the size of the engineering function — but not uniformly. Across the full dataset, 1,360 deployment-spectrum jobs sit alongside approximately 2,400 engineering jobs. But this ratio varies significantly by company type. Pure research labs have almost no deployment roles. Platform companies are SA-heavy. Application companies spread headcount across CSM, SE, and Implementation roles. The aggregate ratio is useful as a headline, but the company-level picture is more nuanced.

What a company sells determines which deployment roles it hires. Platform companies lean on Solutions Architects. Companies with highly customisable products lean on Forward Deployed Engineers. Agent-first companies lean on Forward Deployed PMs. This isn't a single "deployment gap" — it's several distinct gaps shaped by product complexity, customer sophistication, and go-to-market model.

The embedded specialist model is expanding beyond engineering. The Forward Deployed pattern started with engineers and has reached product management — 14 Forward Deployed PM postings exist across five companies, including Sierra (7) and Glean (3). Small numbers, but they follow the early trajectory of the FDE role.

Title standardisation hasn't happened. Six companies in the dataset have invented proprietary titles for roles that map to existing categories. OpenAI alone uses three distinct naming patterns across three lifecycle phases. For anyone evaluating roles in this space — as a candidate or a hiring manager — the job description and the technology requirements are more reliable signals than the title.

The SA/SE boundary may be collapsing at AI companies. The complexity of AI products seems to require both architecture design and hands-on demonstration within the same engagement. Whether these roles formally merge is unclear, but the overlap in our data is significant: 647 combined jobs, with titles and descriptions frequently crossing the boundary.

---

All data is from the Applied Methods dataset as of April 2026. Job counts reflect active postings at time of analysis. The dataset covers 89 AI companies — primarily venture-backed startups and public companies with significant AI operations. It does not cover AI adoption at traditional enterprises. All roles mentioned can be explored at appliedmethods.ai.