AI is creating a new class of jobs that sit somewhere between engineering, product, research, and customer deployment.
Two of the most interesting right now:
Applied AI Engineer
and
Forward Deployed Engineer (FDE).
They sound similar, but they solve different parts of the same problem.
Applied AI makes the AI work better in same environment.
Forward Deployment makes the AI work in the real world in different environment.
The Applied AI Engineer is not usually training a frontier model from scratch. They work where a model meets a real task.
Imagine that a coding agent can complete an impressive demo but fails unpredictably in a large monorepo. The Applied AI Engineer does not merely add a longer prompt. They create a representative task set, trace failures, improve context selection, change the agent’s tool strategy, test regressions, measure latency and cost, and build a loop that turns production failures into better future behavior.
That is almost exactly how OpenAI describes its current Applied AI Engineer, Codex Core Agent role: improving agent behavior on real software-engineering tasks through evals, prompt and tool-use strategies, context construction, production failure analysis and feedback loops.
The finish line is not a successful demo; it is a measurable gain in solve rate, usefulness and economic value.
The recurring work includes:
designing eval datasets that resemble real user work;
building deterministic and model-based graders;
inspecting traces and clustering failure modes;
improving prompts, context, retrieval, memory and tool selection;
deciding when to use prompting, fine-tuning, reinforcement learning or conventional software;
balancing accuracy against latency, inference cost and operational risk;
building guardrails, human-review paths and observability;
turning a one-off experiment into a reusable product capability.
The best Applied AI Engineers think in systems. A model response is only one component. The real product is the complete loop:
input → context → model → tools → action → evaluation → feedback → improvement.
This is also why the role is broader than “prompt engineer.” Prompts matter, but they are only one control surface. Production performance often depends more on the quality of the eval, the tool contract, the available context, the recovery path and the data captured after execution.
Morgan Stanley built an evaluation framework to test AI use cases against expert standards before deployment. Advisors and prompt engineers graded summarization outputs for accuracy and coherence; the team then used those results to refine the system. OpenAI reports that the resulting assistant reached 98% adoption among advisor teams.
The public case study does not disclose the employees’ job titles, so it should not be presented as proof that an “Applied AI Engineer” personally owned the project.
But it is a strong example of the Applied AI work surface: define quality, build representative evals, use expert feedback, improve behavior and scale only when the system earns trust.
The Forward Deployed Engineer begins somewhere else: not with “how do we improve the agent?” but with “what is stopping this organization from changing an important outcome?”
The first answer is rarely “the model.” It may be that data is split across six systems, permissions are unclear, nobody agrees on the acceptance criteria, the existing process contains undocumented human judgment, or the users will not adopt a tool that adds one more screen to their day.
The FDE has to discover that reality and still ship.
Palantir pioneered the title and still describes the role as engineers working side by side with customers, moving from an open-ended operational question to architecture, data work, custom applications and deployment. Its current London role expects 25–50% travel and direct iteration with users.
In frontier AI labs, the role now has a more model-heavy version. Anthropic’s current FDE posting says engineers embed with strategic customers and deliver production artifacts including MCP servers, sub-agents and agent skills.
OpenAI’s FDE team owns discovery, technical scoping, system design, build, rollout and adoption, with travel of up to 50% for its general NYC role.
A typical engagement looks like this:
Sit with operators and observe the real workflow.
Find the decision or bottleneck worth changing.
Define a measurable outcome and launch threshold.
Map data, systems, permissions, risks and human approval points.
Build the smallest end-to-end production path.
Run evals on the customer’s own cases not a generic benchmark.
Deploy, monitor, train users and remove adoption friction.
Hand off the system and convert reusable lessons into product primitives.
The defining skill is not travel. It is context compression: rapidly understanding a domain, a technical stack and an organization well enough to make good decisions before perfect information exists.
OpenAI’s current healthcare FDE role gives unusually concrete detail. The engineer is expected to integrate agentic systems with enterprise APIs, data platforms, EHRs and claims systems; manage PHI, HIPAA, governance and auditability; define human-review and escalation paths; and create customer-specific launch criteria.
The deployment is successful only when it reaches production adoption and measurable workflow impact.
Scale AI’s 2026 work with Mayo Clinic illustrates the kind of problem such teams attack: detecting patient-safety events hidden in routine reporting and reducing administrative work while keeping data inside a secure, HIPAA-compliant environment.
The market does not use a clean taxonomy.
OpenAI’s Applied AI Engineer, Enterprise role is deeply customer-facing. It includes use-case selection, architecture, prototypes, evals, integrations, launch and adoption. Anthropic’s Applied AI Engineer, Enterprise Tech also guides a portfolio of customers from discovery to deployment. Both sound remarkably close to FDE.
The difference appears in emphasis:
Applied AI teams are more likely to own model behavior, eval methodology and reusable technical patterns across multiple customers or a product surface.
FDE teams are more likely to own one strategic deployment end to end, inside the customer’s systems, with a harder commitment to rollout, adoption and handoff.
But this is a spectrum, not a border.
At one end is the product-centric Applied AI Engineer working on Codex agent behavior for millions of users. At the other is an FDE embedded inside a hospital, bank or industrial company. Between them are customer-facing Applied AI Engineers, AI Deployment Engineers, Solutions Architects and Product Engineers.
So when evaluating a role, ignore the title for five minutes and ask:
What is my unit of ownership: a product capability, a portfolio or one deployment?
Who controls the roadmap: the product team or the customer engagement?
Am I accountable for implementation, or mainly for architecture and advice?
How much of my work must become reusable?
Is adoption part of my performance review?
How much travel and on-site work is real—not theoretical?
Those answers reveal the job more accurately than the headline.
Both roles require strong software engineering. Current postings repeatedly ask for Python and often TypeScript or JavaScript, production LLM experience, evaluation frameworks, cloud architecture, security judgment and the ability to communicate with technical and non-technical stakeholders.
But the hiring signal is weighted differently.
You can show an evaluation harness that changed a product decision.
You can explain why an agent failed and how you improved it systematically.
You understand prompting, retrieval, tool use, fine-tuning and their trade-offs.
You have shipped feedback loops using production traces or user judgments.
You can discuss quality, latency, cost and safety as one optimization problem.
You owned a deployment from ambiguous discovery to production adoption.
You integrated multiple enterprise systems, permissions and data sources.
You can translate a business workflow into measurable technical acceptance criteria.
You have earned trust with operators, executives, security teams and engineers.
You know when to build a reusable primitive and when to ship a pragmatic customer-specific solution.
Former founders often fit the FDE profile because they already combine product discovery, coding, delivery, sales empathy and extreme ownership. Technical Product Managers can also transition, but the market is explicit: stakeholder management without substantial personal contributions in code, architecture, evals or debugging is not enough.
Choose Applied AI Engineer if you enjoy asking:
Why did the model fail?
How do we measure “good”?
Which data or feedback will improve the system?
How do we make this reliable and reusable for many users?
Choose Forward Deployed Engineer if you enjoy asking:
What is the real operational problem?
Which constraint is blocking value?
What can we ship in four weeks inside this environment?
How do we get people to adopt it and make the customer independent?
Applied AI offers deeper continuity with a product and its intelligence. FDE offers broader exposure to industries, workflows and executive decisions. Applied AI can lead toward Staff AI Engineer, AI Product Engineering or research-engineering roles. FDE can lead toward deployment leadership, product, solutions leadership, entrepreneurship or a domain-specific AI company.
Neither role is “more technical.” Their technical depth points in different directions.
The weak Applied AI Engineer produces clever demos that cannot be measured. The weak FDE produces bespoke consulting work that never compounds. The great version of each role avoids both traps.
OpenAI — Applied AI Engineer, Enterprise
Location: San Francisco, New York City, or US remote.
Published compensation: $197k–$278k + equity.
Role: https://openai.com/careers/applied-ai-engineer-enterprise-san-francisco/
OpenAI — Applied AI Engineer, Codex Core Agent
Location: San Francisco, New York City, Seattle, London.
Published US compensation: $230k–$385k + equity.
Role: https://openai.com/careers/applied-ai-engineer-codex-core-agent-san-francisco/
Anthropic — Applied AI Engineer, Enterprise Tech
Location: San Francisco, New York City, Seattle.
Published compensation: $200k–$320k.
Role: https://job-boards.greenhouse.io/anthropic/jobs/5057647008
Scale AI — Applied AI Engineer, Global Public Sector
Location: London, Doha.
Requirements: 7+ years in engineering and 2+ years working with production AI/ML systems.
Compensation range is not published.
Role: https://scale.com/careers/4413992005
Scale AI — Staff Applied AI Engineer
Location: London.
Senior role focused on evaluation, governance, and reusable AI capabilities.
Compensation range is not published.
Role: https://scale.com/careers/4720050005
OpenAI — Forward Deployed Engineer
Location: New York City.
Published compensation: $162k–$280k + equity.
The role can involve up to 50% travel.
Role: https://openai.com/careers/forward-deployed-engineer-%28fde%29-nyc-new-york-city/
OpenAI — Forward Deployed Engineer, Healthcare
Location: San Francisco.
Published compensation: $198k–$280k + equity.
OpenAI asks for 6+ years of experience, with up to 50% travel.
Role: https://openai.com/careers/forward-deployed-engineer-%28fde%29-healthcare-sf-san-francisco/
Anthropic — Forward Deployed Engineer
Location: San Francisco, New York City, Seattle.
Published compensation: $280k–$320k.
Approximately 25% travel.
Role: https://job-boards.greenhouse.io/anthropic/jobs/5302966008
Palantir — Forward Deployed AI Engineer
Location: London.
Focused on GenAI strategy and implementation with customers.
Can involve up to 25% travel.
Compensation range is not published.
Role: https://jobs.lever.co/palantir/ff1029bd-bb6d-4d78-a03e-5f9744d0b798
Palantir — Forward Deployed Software Engineer
Location: London.
Customer-embedded, full-time hybrid role with roughly 25–50% travel.
Compensation range is not published.
Role: https://jobs.lever.co/palantir/5168e8fd-fec1-4fea-b7a1-81bdaea65850
Palantir — Forward Deployed Software Engineer Internship
Location: Paris.
On-site internship for students graduating in 2027.
Role: https://jobs.lever.co/palantir/1b6f1d82-d459-4dea-8bc2-8d2ffe6f881a
OpenAI — Applied AI Architect, Digital Natives
Location: Paris.
This is an adjacent role focused on technical account strategy, architecture, and helping customers move AI systems into production.
Role: https://openai.com/careers/applied-ai-architect-digital-natives-paris-france/
In the SaaS era, software companies could ship a product and leave implementation to customers, partners or consultants.
AI is different. The product is probabilistic. Quality depends on the customer’s data, workflow, tools, permissions, policies and expert judgment. A model can be generally intelligent and still fail at the last mile.
This is why Applied AI and Forward Deployment are becoming core product functions rather than support functions.
The Applied AI Engineer closes the gap between capability and reliability.
The Forward Deployed Engineer closes the gap between reliability and adoption.
The companies that learn fastest will connect both loops: failures in the field become evals; evals become product improvements; product improvements accelerate the next deployment.
That feedback system not the job title is the real competitive advantage.
And that may be the defining engineering organization of the AI era: not research, product and customers in separate buildings, but one continuous loop from frontier model to real-world outcome and back again.
If this was useful, subscribe for more deep dives on AI-native product roles, agent infrastructure and how software companies must change for the agent era.
No posts
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.