How to Become a Forward Deployed Engineer: A Practical Guide
A step-by-step guide to becoming a forward deployed engineer: finding businesses, mapping workflows, and deploying AI agents to automate them.
What does a forward deployed engineer actually do?
A forward deployed engineer (FDE) walks into a business, figures out which of its processes can be handed off to AI agents, and builds the systems to do it. Instead of writing software that a company deploys itself, the FDE embeds directly with department heads or owners, maps out how work actually gets done, and automates the parts that don’t need a human. The role is sometimes called “agent operator,” and it has become one of the most in-demand jobs tied to the current wave of AI capability, because it sits at the bottleneck: companies have access to the same models everyone else does, but very few people know how to turn those models into working automation inside a real organization.
TL;DR
- Forward deployed engineers go inside businesses, interview the people doing the work, and convert processes into tasks that AI agents can run end to end.
- Pay is already high: median total compensation in the US reportedly sits around $200,000 to $250,000 a year, with top performers near $400,000, and Anthropic has posted individual roles in the $300,000 to $350,000 range.
- Anthropic is investing heavily in the pipeline, reportedly committing hundreds of millions of dollars to train thousands of these engineers by the end of 2027, and calling the talent the hardest to find.
- The job splits into two buckets: tasks an agent can do completely on its own (huge, fast uplift) and tasks that need human judgment (slower, more collaborative upskilling work).
- Small and mid-size businesses usually need help on the growth side of the business, not the back office, which is why FDEs often focus on reach, acquisition, conversion, and expansion.
- The toolkit is simple: a coding agent like Claude Code or Codex, a way to record or transcribe how people work, and the discipline to turn that into a repeatable standard operating procedure.
- The actual bottleneck isn’t the automation itself, it’s getting into businesses and getting people to clearly explain how they work.
Nothing gets lost.
Why is this role suddenly valuable?
Pay in any economy moves toward whatever is scarce. For years, the ability to write software was scarce, so programmers were paid well. As AI models got good at writing code, that scarcity disappeared, and so did much of the pay premium for pure coding work. The same logic applies to other fields: doctors and lawyers are compensated highly in part because demand for their judgment outstrips supply.
Forward deployed engineering sits in a different spot. The underlying technology, the models themselves, is now abundant and cheap to access. What’s scarce is the ability to go into a real business, understand its actual workflows, and translate that mess into something an agent can execute reliably. That translation work, part consulting, part systems engineering, part change management, is the current bottleneck. Anthropic has been explicit about this, describing forward deployed talent as difficult to find and backing that claim with a large, multi-year training investment aimed at producing thousands of these engineers by 2027.
There’s also a capability trend driving urgency. Independent research groups that track how long AI agents can work autonomously while maintaining human-level quality have found that this “time horizon” has been doubling roughly every several months. Newer models can already operate unsupervised for extended stretches on real work inside AI labs. As that trend continues, the amount of economic value a single person can direct through agents keeps climbing, which is exactly why operating agents well, rather than performing the underlying labor yourself, is becoming the valuable skill.
How do you find and map the work an agent can automate?
The process starts with conversations, not code. An FDE typically:
- Gets access to the people doing the work. That means department heads, process owners, or the business owner directly if it’s a small or mid-size company.
- Has them walk through their actual process, step by step, not the idealized version written in a manual somewhere.
- Captures that process on record. This can be as simple as shadowing someone and recording their screen, taking periodic screenshots of their digital work, or using an audio transcription device during meetings to capture everything said.
- Feeds that raw log into an AI model to turn it into a structured standard operating procedure (SOP). Once a process is written as a clear SOP, it becomes something an agent can attempt to execute, using browser automation or computer-use capabilities to click, type, and navigate the way a person would.
- Sorts every task into one of two buckets: fully agent-capable (no human needed) or collaborative (human judgment still required).
Nothing gets lost.
For agent-capable tasks, the entire task gets handed to the agent. Because these tasks can be done with no human bottleneck at all, the productivity gains can be enormous, sometimes described as several thousand percent uplift against the old manual process. For collaborative tasks, the goal shifts: instead of replacing the person, the FDE trains them to use an agent as a thought partner, which still produces meaningful efficiency gains, just smaller ones, often in the range of a few hundred percent rather than several thousand.
Where should you focus first inside a business?
Most small and mid-size organizations, roughly under 100 employees, don’t struggle with their back office as much as they struggle with growth. Their constraint usually sits on the front end: generating leads, closing deals, and expanding existing accounts. A useful framework for this is “RACE”: reach, acquisition, conversion, and expansion. Forward deployed engineers who focus on RACE processes tend to find the clearest, fastest wins, because growth bottlenecks are usually visible, measurable, and painful enough that an owner will prioritize fixing them.
A concrete example: a home services company that’s bottlenecked on generating proposals or quotes quickly enough to keep up with inbound leads. An FDE might build a system that uses AI to draft a first-pass proposal automatically, triage incoming requests by urgency, or queue and summarize emails every morning. None of this requires reinventing the business, it requires understanding exactly where time is being lost and routing that step to an agent.
What tools do you actually need to do this work?
The technical setup is intentionally lightweight. A coding agent like Claude Code or OpenAI’s Codex, both available as roughly $20-a-month subscriptions, can handle most of the build work: writing scripts, wiring up integrations, and automating browser-based tasks. Some practitioners run multiple agents in parallel using terminal-based orchestration tools so they can delegate several build tasks at once instead of babysitting one agent at a time.
The day-to-day workflow often looks less like traditional engineering and more like project management: talking to a business owner or department head, capturing what they need in plain language, and handing that context to an agent to build. Much of the actual “coding” happens by describing a need conversationally and letting the agent generate the system, whether that’s a Python script that triages a Gmail inbox every morning or a tool that auto-generates sales proposals from a template.
The real constraint isn’t the build. Current models are capable enough to construct most of these systems quickly. The constraint is access: getting into a business, earning enough trust that people will walk you through their actual process, and getting them to commit the time to do it properly.
Is becoming a forward deployed engineer worth it?
If you’re evaluating this as a career path, the economics look favorable right now. Compensation is already well above typical software roles, demand is backed by public commitments from major AI labs to train more people into the role, and the skill set (process mapping, client communication, and agent orchestration) is harder to automate than pure coding, since it depends on judgment and relationship-building that current models can’t fully replace. The tradeoff is that it’s not a purely technical job. Success depends heavily on your ability to get access to decision-makers, extract clear information from people who aren’t used to documenting their own workflows, and manage the human side of replacing or upskilling roles inside a company.
Frequently Asked Questions
What is a forward deployed engineer?
A forward deployed engineer is someone who works directly inside a client business to identify processes that AI agents can automate, then builds and deploys the systems that do it, rather than building generic software for broad distribution.
How much do forward deployed engineers get paid?
Reported figures put median total compensation in the US around $200,000 to $250,000 a year, with top earners near $400,000. Anthropic has advertised individual roles in the $300,000 to $350,000 range.
Do I need a coding background to become one?
You need enough technical fluency to work with coding agents like Claude Code or Codex and to understand what’s feasible to automate, but the core differentiator is the ability to extract and map business processes, not deep software engineering expertise.
What’s the difference between an FDE and a general AI consultant?
The forward deployed model emphasizes embedding directly within a specific business’s workflows and shipping working automation, rather than giving strategic advice from the outside. The output is a running system, not a recommendation.
Why are companies like Anthropic investing in training these engineers?
Because the bottleneck in deploying AI effectively isn’t model capability, it’s the scarcity of people who can translate messy real-world business processes into tasks an agent can reliably execute. That shortage is why labs are funding training pipelines for this specific role.
