Forward Deployed Engineer: OpenAI Launched Today. Anthropic Launched First. The Headlines Missed Who Just Won.
May 13, 2026·12 min read
TL;DR
Two frontier labs put $5.5B behind FDE businesses in 8 days. Goldman's Marc Nachmann named the bottleneck: 'democratize access to forward-deployed engineers.' The labor market doubled in 14 days — 187 → 399 LinkedIn JDs, AI/ML requirement 71% → 80%, salary disclosure 11% → 55%. Here's the complete 2026 read on the role: skills, salary, path in, and the explicit answer to 'should I go for AI Engineer or FDE?'
OpenAI launched today. Anthropic launched first. The headlines missed who just won.
OpenAI's vehicle is called The Deployment Company — $4B for forward-deployed engineering, backed by TPG, Advent, Bain Capital, and Brookfield. Anthropic's, launched eight days earlier on May 4, is a $1.5B enterprise AI services venture with Blackstone, Hellman & Friedman, and Goldman Sachs. Most of the LinkedIn feed today is reposting one of those two announcements as if they're separate stories.
They're the same story.
The line that matters more than either announcement came from Goldman's Marc Nachmann, sitting on the Anthropic deal: the new venture's purpose was to "democratize access to forward-deployed engineers."
Translation: the labs aren't capital-constrained. They are FDE-constrained. The bottleneck is one specific type of engineer — and the headlines today are naming everything except who.
This piece is the full read on that role: what it actually is, what the skill demand looks like across 399 fresh JDs, what the salary band is, the explicit decision logic between AI Engineer and FDE if you're torn between the two, the fastest path in by background, and what to do this week if you're in the window.
What 14 days of JD data confirms
I have been tracking the Forward Deployed Engineer market for the last year — the rolling JD scans, the cohort conversations, the AMA questions from working engineers. Two weeks ago, I posted the first FDE scan: 187 JDs across the US market, $150K–$325K, 71% requiring AI/ML, the role spreading fast but without a stable definition across companies.
I re-ran the scan this week to confirm what the announcements suggested.
399 JDs, up from 187. The pool doubled in 14 days.
80% require AI/ML, up from 71%. The role is consolidating around AI deployment specifically. The remaining 20% are mostly older "Forward Deployed" postings that haven't refreshed yet.
Compensation disclosure jumped to 55%, up from 11%. Companies are competing on transparency now — which happens when the role has enough definition that hiring managers know what to pay.
New names on the hiring list. OpenAI. Anthropic. Notion. Vercel. Snorkel AI. Together AI. Glean. Cohere. Reflection. Plus Deloitte alone is running 70 simultaneous FDE postings. Google is at 20.
The 8-day frontier-lab convergence and the 14-day labor-market doubling are the same signal at two altitudes. The labs named the function. The hiring data confirmed it.
The convergence is wider than two announcements
OpenAI and Anthropic got the headlines. They are not the only labs moving.
- Google DeepMind — Forward Deployed AI Engineers embedded via Accenture, Capgemini, Deloitte, HCLTech, PwC, TCS for Gemini Enterprise (Cloud Next 2026, April). Citi Wealth's "Citi Sky" agent is the named case study.
- Databricks — productized "AI FDE" team listed by that exact name on its careers page. Not "applied AI." Not "customer engineering." AI FDE. The category gets the title.
- Cohere + Mistral — both hiring "Applied AI / Forward Deployed ML Engineer" roles across US and EMEA in 2026. The phrasing varies. The function does not.
- Accenture — "Reinvention Deployed Engineers" (~30K Claude-trained consultants) — Accenture's house name for the FDE archetype.
- EPAM — 250 "FDE Black Belts" inside a 10K-architect Anthropic practice.
CIO Magazine called FDEs "the new AI limiting factor" on May 6, citing a Gartner prediction that 70% of enterprises will abandon agentic-AI deployments because of FDE supply, not technology.
This is what a category looks like when it stops being a hiring experiment and starts being a market.
What the labs are actually bidding for
The labs are not bidding for "AI engineers." They are not bidding for "senior SWEs." They are bidding for the rare engineer who carries BOTH technical depth AND customer-facing experience. The overlap of two skills most engineers spent the last decade picking between.
SWEs and MLEs went technical depth. PMs and solutions roles went customer-facing. The small overlap — the engineers who got told "too technical for PM, too customer-facing for engineering" — that is exactly who the labs are buying. That combination is the entire moat.
If that's you, here's what just happened to your market:
Low competition — because both signals on one resume is rare. The supply side is thin precisely because most engineers don't carry both. That thinness is structural, not temporary.
High opportunity — because $5.5B of PE capital just admitted FDE supply is the bottleneck. Two frontier labs. The investor side naming it out loud. The labor market doubling its postings in 14 days. All four pieces of evidence point at the same engineer profile.
This is the strongest hiring window the market will give you in 2026. It is also the narrowest.
What FDE actually means
The title is newer than the work. Forward Deployed Engineers have been showing up in AI companies for years — they were just called different things. Now the category has a name, a salary band, and (as of this week) productized billion-dollar businesses around it.
The defining characteristic: you are the technical owner inside the client's environment, not your company's codebase. If the deployment fails, that's on you.
How FDE differs from adjacent roles people confuse it with:
vs Software Engineer. SWE builds product in-house and hands it off. FDE deploys inside the client's environment and owns outcomes there. You're not building a product; you're making a product work in a specific, messy real-world context.
vs Sales Engineer. SE supports the deal. FDE lives post-deal. Accountable for the thing actually working in production after the contract is signed.
vs Solutions Architect. SA designs the blueprint and hands it over. FDE owns the full implementation end-to-end and stays accountable until it ships.
vs PM. PM owns the roadmap. FDE owns the technical integration inside a specific customer environment. The scope is narrower, the hands-on bar is higher.
"Without FDEs, we risk having thousands of customers stuck in pilot purgatory — signed up, but not successfully deployed." — Sarah Khalid, FDE Director, Salesforce
That's the job in one sentence. The FDE exists because enterprise AI doesn't deploy itself.
FDE vs AI Engineer
This is the section most readers are here for. Both roles write production-grade code daily. Both work with AI/ML systems. The difference is where and for whom.
AI Engineers build features inside their own company's product. FDEs deploy AI systems inside the client's environment — different data, different infrastructure, different stakeholders, and a delivery clock that resets with every new client. AI Engineers optimize for the product. FDEs optimize for the client's specific outcome.
| FDE | AI Engineer | |
|---|---|---|
| You own | AI systems in client production | AI features in your product |
| Customer exposure | Deep — weeks to months onsite | Minimal — internal teams only |
| Core skill mix | Technical depth + client delivery — simultaneously | Technical depth |
| Interview | Coding + systems design + client scenario | Coding + systems design |
| Salary (May 2026) | $150K–$325K (399 JDs) | $153K avg (Indeed) |
| Best path in | SWE / DS / MLE / Solutions / TPM with client-facing experience | Backend / Full-stack SWE |
| Travel / onsite | Often required | Rare |
Who should go for AI Engineer:
- You love heads-down product engineering. Long arcs inside one codebase energize you.
- You want to ship features and iterate fast on the same product over months.
- Customer interaction is draining or distracting, not energising.
- You optimise for technical depth — internals of frameworks, model architectures, performance tuning.
- Your background: strong backend, full-stack, or ML-internals engineering.
Who should go for FDE:
- You're energised by customer problems and ambiguous delivery contexts. The messy half of integration is the part you find interesting, not the part you avoid.
- You enjoy translating between non-technical stakeholders and production-grade systems.
- You can hold accountability under client pressure — including when the AI does something stupid and you're the one explaining it.
- Your background already shows the both-sides signal: SWE who worked on customer-facing projects; DS or MLE with consulting / client delivery; solutions architect with deep technical ownership; technical PM with real implementation history.
Torn between the two? The fastest disambiguator: think about the last engineering project you were proud of. Was the proudest part the architecture, or was it the moment you got the customer's nod that "yes, this is what we needed"? AI Engineer is the first. FDE is the second.
What the 399 JDs actually ask for
Refreshed from the May 12 scan.
Core technical skills:
| Skill | % of JDs | Notes |
|---|---|---|
Python |
51% explicit (~80% baseline) | Frequently unstated — assumed baseline |
| APIs & system integration | 17% | Integration architecture depth |
| AWS / cloud (GCP, Azure) | 14% explicit | Multi-cloud fluency preferred |
| Data engineering / SQL / pipelines | 13% | Becoming essential for client-side deployments |
| TypeScript / React | 9% React explicit | Front-end for client tooling and deployment visibility |
Kubernetes / Docker / Terraform |
10–15% combined | Infrastructure provisioning baseline |
AI/ML skill clusters (80% of JDs require at least one):
| Cluster | JD coverage | What employers want |
|---|---|---|
| Agentic AI development | 150 JDs | Multi-agent systems, LangGraph, CrewAI, AutoGen, agent orchestration, agent reliability |
| AI/ML system integration & deployment | High | MLOps / LLMOps, LLM enterprise integration via APIs, MLflow, scalable AI systems |
| AI/ML evaluation & optimization | High | Model monitoring, prompt management, evals, fine-tuning, RAG pipelines, observability |
| Emerging GenAI / LLM tech | High | Bedrock, Databricks AI Gateway, vector DBs, MCP, Claude Code, Responses API, OpenAI Agents SDK |
Trajectory note: AI/ML isn't an add-on for FDE — it's the core. The 80% requirement (up from 71% in April) reflects the role consolidating into primarily an AI-deployment function. If your JDs from a year ago didn't mention LangChain or agent orchestration, the current ones almost certainly do.
Self-assessment heuristic: Read 10–15 current FDE JDs. If you can read the body paragraphs without needing to look up more than 2–3 terms per JD, you are roughly inside the 80%. If you need to look up more than 5 terms per JD, the gap is real but closable in 4–8 weeks of focused shipping.
Salary and hiring landscape
55% of JDs now disclose salary (up from 11% in April), so this band is the most accurate read of the role's compensation at any point so far.
| Level | Range | Disclosed examples (May 2026) |
|---|---|---|
| Mid | $113K–$208K | Tagup: $130K–$160K |
| Senior | $137K–$282K | Snorkel AI: $172K–$300K · Red Hat: $184K–$305K · Kore.ai: $180K–$230K |
| Lead / Manager | $200K–$325K | David Joseph & Co: $175K–$250K · Acceler8 Talent: $150K–$250K + equity |
The 55% disclosure rate is itself a signal. Companies disclose salary when they need to compete on transparency to attract candidates — which happens precisely when supply is constrained. Treat the bands as floor-leaning rather than ceiling — top-tier AI shops will pay above, often with equity that significantly changes total comp.
Largest-volume hirers (May 2026):
| Company | JDs | Salary (if disclosed) |
|---|---|---|
| Deloitte | 70 | $113K–$282K |
| 20 | $153K–$300K + bonus + equity | |
| Tagup | 4 | $130K–$200K |
| Acceler8 Talent | 4 | $150K–$250K + equity |
Other notable names in the May 2026 hiring pool: OpenAI, Anthropic, Scale AI, Tesla, Cohere, Notion, Vercel, Cursor, HeyGen, H2O.ai, Labelbox, Unstructured, SuperAnnotate, StackAI, Snorkel AI, Together AI, Glean, Reflection, Mistral, LiveKit, Arcade.dev, Parloa, Kore.ai, Seekr, plus 100+ others.
When Deloitte runs 70 simultaneous FDE postings, the work has moved past pilot stage at their clients. They are now staffing for sustained deployment volume. That's a Big Three consulting firm building a P&L line around the role.
The interview format
Three components, based on practitioner signals:
1. Coding round. Standard SWE coding — LeetCode-style or practical implementation. Doesn't need to be exceptional; needs to clear a competency bar. The interview isn't primarily testing algorithm optimization.
2. Systems design. Architecture and integration design, increasingly AI-flavored. Expect prompts like: design an LLM-powered workflow for a client use case, or architect an AI observability system for an enterprise deployment. Know how to design with LLM APIs under constraints: latency, cost, reliability, fallback handling.
3. Client scenario. A consulting-style case. "The client's deployment broke after go-live. Walk me through how you handle it." Tests communication under pressure, problem decomposition, and whether you can hold accountability without blaming the client or the product.
The round-three filter. The client scenario round is the round that filters out engineers who've never worked client-facing. Pattern recognition for what to escalate, when to push back, and when to just fix it — that's what's being tested. Practitioners report this round catches more candidates than the technical ones.
Practitioner signal worth flagging: OpenAI API design questions are appearing in systems design rounds across the May 2026 hiring pool. Know how to architect with LLM APIs, not just use them.
The fastest path in, by background
The fastest path in is not deeper pure engineering. It's combining technical credibility with a customer-facing track record.
| Background | What carries over | The gap to close |
|---|---|---|
| Backend / Full-stack SWE | APIs, system design, deployment, Python — ~70% of the stack |
Client-facing experience + AI/ML skills (LLM APIs, RAG, agents). One deployed client integration project closes most of the gap. |
| DS or MLE with client delivery | Model deployment, ML fundamentals, Python |
Client delivery framing on the resume. If you've deployed models in client environments, you're already doing the job — the gap is mostly narrative. |
| Solutions Architect | System design thinking, stakeholder communication, enterprise architecture | Implementation depth. SA designs what FDE builds. The gap is hands-on delivery ownership, not technical concepts. |
| Technical PM with implementation history | Product and stakeholder communication, requirements translation | Hands-on delivery credibility. The bar is that "implementation history" is real — not just oversight or ticket-writing. |
No PhD-level math prerequisite. FDE interviews test systems thinking, client scenario handling, and code quality — not model mathematics or probability theory. The technical bar is production-grade engineering, not academic ML depth.
The one non-negotiable signal regardless of background: evidence that you deployed something in a client's environment and owned the outcome. Not contributed to. Not designed. Owned, shipped, and were accountable for it working.
What to surface on your resume and LinkedIn:
- Client environments you've deployed to (not just internal systems)
- End-to-end ownership language — "owned" not "contributed to"
- AI/ML work that shipped to production, not just modeled or prototyped
- Communication with non-technical stakeholders under delivery pressure
What a strong FDE portfolio looks like
FDE portfolios are fundamentally different from standard engineering portfolios. Hiring managers want evidence that you can own end-to-end delivery, work with real constraints, and communicate outcomes to non-technical stakeholders. Technical depth alone doesn't signal this.
Weak vs strong FDE portfolio:
| Weak (avoid) | Strong (target) |
|---|---|
| Pure technical projects with no customer context | Customer problem → solution → measurable impact |
| Backend or ML only, no user-facing | Full-stack with demo-ready interface |
| Long development cycles (months to build) | Rapid prototypes built in days or weeks |
| Polished but isolated systems | Messy integration with real constraints (legacy APIs, security, bad data) |
| No documentation of trade-offs | Clear articulation of what you chose and why |
| Solo projects only | Evidence of working across technical and non-technical stakeholders |
3–5 projects to ship:
- Client-style POC — end-to-end solution for a simulated business problem with a working demo.
- Integration project — connect an AI system to a legacy API, messy database, or enterprise tool.
- Customer-facing AI app — chatbot or agent with a non-technical user interface (Streamlit, Gradio).
- Data pipeline + executive dashboard — ETL from messy sources → analysis → output a non-technical stakeholder can read.
- Technical case study — written breakdown: the constraint, your approach, the outcome, what you'd do differently.
The signal hiring managers are actually looking for: can you operate like a startup CTO dropped into a client's environment — owning the problem, adapting to what's there, and shipping something that works? GitHub repos where the README explains the business problem, not just the code. Demo videos or live apps.
Who FDE is NOT for
The role has real downsides and is wrong-fit for a meaningful slice of engineers.
Engineers who want to stay heads-down in the codebase. Client interaction isn't optional — it's the core of the job. If customer-facing work feels like a distraction, this will be a bad fit regardless of technical ability.
People who find ambiguity and delivery pressure draining. Client deployments are rarely clean. Requirements change mid-project, integrations break in production, and the stakeholder on the client side may not understand what they asked for. The job requires tolerating that without spiraling.
Engineers early in their career without deployment experience. This role requires owning outcomes, not just writing code. Without prior deployment accountability, you won't have the pattern recognition for when to escalate, when to push back, and when to just fix it.
PMs without hands-on implementation history. The technical bar is real. Stakeholder management experience doesn't substitute for it.
If two or more of these describe you, the AI Engineer track (or remaining in your current role and deepening) is the higher-leverage move.
Three reads, by where you sit
If you got cut six months ago and are still looking. This is the seat most likely to reopen at your level — it did not exist as a named role when you last interviewed. The 80% AI/ML bar is the screen. One agentic project shipped publicly in the gap — with a README that documents what the AI got wrong and how you caught it — closes most of it. Compensation floor at mid level: $113K. Ceiling at senior: $282K. Six months ago, you would not have known either number with this confidence.
If you survived the cut at your current company. Your senior SWE title and an FDE title now read approximately 70% the same JD, with one delta: client deployment ownership. The lateral window is wider this quarter than it has been at any point in the role's existence. The pool of qualified candidates is small enough that hiring managers are still flexible on background; the pool of engineers who know the role exists is also small enough that supply hasn't caught up. Both shrink in six months.
If you are a solutions engineer, TPM, or consultant with technical chops. The market is asking for exactly your shape. The "non-traditional" background you've apologised for on resumes is now the JD. Don't lead with "I'm not really an engineer" — lead with "I've owned delivery in client environments." That's the screening signal.
What to do this week
Three actions, in order.
1. Pull the JDs. Go on LinkedIn. Search "Forward Deployed Engineer" plus your region or timezone. Read 10–15 postings in full. Not the headlines. The body paragraphs. The "preferred" sections. You are looking for the line that recurs in two out of three — that phrase is the actual screening signal. In May 2026 it is almost always agentic tooling fluency, LLM-deployment experience, or evals.
2. Name the gap. Two columns. Left: what your last six months of work demonstrates clearly. Right: what the JDs are explicitly asking for. The diff is what you ship in the next four weeks. The diff is almost always smaller than people expect — usually one project, not a curriculum.
3. Write one public artifact. A project, a blog, a README. Document the deployment context, the constraint that mattered, the call you made when the AI failed in production-shaped conditions. That artifact is what gets past the round-three client-scenario filter — every candidate who passes that round has a version of this.
Why this update exists
OpenAI productized the function on May 11. Anthropic did it eight days earlier. The Goldman quote named the bottleneck. The labor market doubled its postings in fourteen days. The credentialing layer is forming.
Two weeks ago, I called the two-skill engineer profile "the job description." This week, $5.5B of PE capital paid for exactly that. The thesis around when to move has changed in eight days, twice.
The macro signal is unambiguous. The role is named, the salary is publicized, the frontier labs have built productized businesses around it, and the supply side is publicly thin enough that an investor at Goldman Sachs is using the word democratize to describe access. If you fit the profile — technical depth + customer-facing experience — this is the cleanest hiring window of 2026.
Source: LinkedIn FDE JD scans — 187 (Apr 28, 2026) refreshed to 399 (May 11–12, 2026, US market, 30-day rolling, mid through lead level). PE capital deals: Anthropic enterprise AI services venture (May 4, 2026) · OpenAI Deployment Company (May 11, 2026). Goldman Sachs commentary (Marc Nachmann) via Fortune, May 4, 2026. CIO Magazine, May 6, 2026. Google Cloud Next 2026, Accenture–Anthropic and EPAM–Anthropic partnerships. Sarah Khalid quote — Salesforce FDE leadership, 2025. · Dexity.com
Dexity Sprint
Ship Production Code with AI
Most senior engineers have tried Cursor or Claude Code and ended up with larger PRs, more review cycles, and hidden technical debt.
