Where did the term come from?
The phrase is borrowed from Palantir, which built its business on engineers who deploy into a customer's organization and deliver outcomes inside that customer's systems, not from a vendor booth. The model spread because it solved a specific failure: software that works in a demo and dies in the customer's real environment.
In May 2026 the term went mainstream. OpenAI and Anthropic each launched a deployment subsidiary within a day of each other, both using the title Forward Deployed Engineer, both aimed at getting frontier models into enterprise products rather than selling API access and hoping. Adding AI to the title marks the shift: the capability being deployed is no longer just software, it is an AI workflow.
What does a Forward Deployed AI Engineer actually do?
They take one narrow use case from a vague idea to a working AI feature inside the customer's product, end to end, and they own every step in between.
The work is concrete, not advisory. A typical engagement runs through:
- Scoping. Turn "we should add AI" into one workflow with a named user, a clear input, and a clear output.
- Architecture and UX. Decide where the workflow lives in the product and how a person actually triggers and reviews it.
- Prompt and context design. Build the model interaction, including the system prompt, the retrieved context, and the guardrails.
- Data integration. Connect the workflow to the customer's real data and systems, often through the Model Context Protocol.
- Implementation. Write production code in the customer's repository, with edge cases, fallbacks, error states, and logging.
- Evaluation. Ship a written rubric that defines what good output looks like and how the team scores it over time.
- Handoff. Hand the team code they can maintain, plus the rubric and a short playbook, so the workflow survives without the engineer.
The output is code in the customer's repo. Not a dashboard. Not a report. Not a Loom.
How is a Forward Deployed AI Engineer different from a solutions engineer or ML engineer?
All three are technical, but they ship different things to different places. A Forward Deployed AI Engineer ships an AI workflow into the customer's own product and owns the outcome.
| Role | What they ship | Where the work lives | Optimizes for |
|---|---|---|---|
| Forward Deployed AI Engineer | A working AI workflow built into the customer's product | The customer's codebase and product | A shipped outcome the customer's team can run and maintain |
| Solutions engineer | Integrations, demos, and configuration of a vendor's product | The boundary between vendor and customer | Closing and onboarding the customer onto the vendor's platform |
| ML engineer | Models, training pipelines, and inference infrastructure | The model and data platform | Model performance, scale, and reliability |
| AI consultant | Strategy, roadmaps, and recommendations | Documents and meetings | Advice and direction, with execution left to others |
For a side-by-side that goes deeper on the closest comparison, see Forward Deployed AI Engineer vs solutions engineer.
What skills does the role require?
Less model research, more product engineering and judgment. The hard part is usually deciding what to build and proving it works, not calling the API.
- Product engineering. Shipping real features into a real codebase, with the discipline that production demands.
- Prompt and context design. Getting reliable output from frontier models, including retrieval and structured context.
- Data integration. Connecting models to messy customer systems and APIs, often via MCP.
- Evaluation. Defining what good output is and scoring it, so quality is measured rather than asserted.
- Customer-facing judgment. Scoping under pressure, saying no to the wrong workflow, and protecting a fixed timeline.
How do you become a Forward Deployed AI Engineer?
Most people who do this well arrive from product engineering, solutions or consulting, or a founder background, then prove they can ship one narrow AI workflow end to end.
The fastest credible path is to build and ship a real workflow, write the evaluation rubric for it, and be able to walk someone through the code. For a step-by-step version, see how to become a Forward Deployed AI Engineer.
Forward Deployed AI Engineering for founder-led B2B SaaS
The enterprise version of this role is built for Fortune 500 procurement. The same work, sized for a 5-to-50-person founder-led SaaS team, is the AI Feature Sprint: one narrow AI workflow shipped into the product in 10 business days, fixed scope and fixed price.
That is the work I do. Every pattern I ship into a customer's codebase has already shipped in Vectig, my own AI-native product and live R&D lab. If you want to see the engagement, start with the AI Feature Sprint or a Discovery Day.
The book
A working guide to the role
Forward Deployed AI Engineering: A Working Guide to the Hottest Job in Software is the long-form version of this page: what the role is, where it came from, how it is practiced, and how to do it well. Written from inside the work, in the months after OpenAI and Anthropic put the title on the map.
About the author
Frequently asked questions
Is a Forward Deployed AI Engineer the same as a forward deployed software engineer?
It is the same delivery model applied to AI work. A forward deployed software engineer embeds with a customer and ships software into their systems. A Forward Deployed AI Engineer does the same, but the capability shipped is an AI workflow: prompt design, model integration, data plumbing, and an evaluation rubric, built into the customer's product.
Do you need a machine learning background to be a Forward Deployed AI Engineer?
No. The role is about applying existing models, not training them. The core skills are product engineering, prompt and context design, data integration, evaluation, and customer-facing judgment. A research or MLOps background helps in some engagements but is not the entry requirement.
Is Forward Deployed AI Engineering only for large enterprises?
No. Palantir, OpenAI, and Anthropic run forward deployed teams aimed at Fortune 500 procurement, but the same shape of work fits founder-led B2B SaaS companies of 5 to 50 people. The engagement is just sized down: one narrow workflow, fixed scope, shipped in days rather than quarters.
What tools does a Forward Deployed AI Engineer use?
Frontier models from providers like Anthropic and OpenAI, an agentic coding tool such as Claude Code, the Model Context Protocol (MCP) to connect models to the customer's systems, and a written evaluation rubric to keep output quality measurable. The customer's own stack is the deployment target.
How is this different from AI consulting?
A consultant delivers advice: a strategy, a roadmap, a recommendation. A Forward Deployed AI Engineer delivers working code in the customer's repository. The judgment is packaged with the execution. The deck, if there is one, describes something that already runs in production.
Newsletter
Notes from the edge of building
One email a week on Forward Deployed AI Engineering, building AI-native SaaS solo with Claude Code, and the unglamorous middle of shipping.
No spam. Unsubscribe in one click.