AI & careers · Field notes
What Is a Forward Deployed Engineer—and Why Does It Matter Now?
Understand the engineer who turns a customer’s problem into working software, why AI makes that work more relevant, and the skills to start practicing.
- Written
An AI demo can look great and still leave the business with the same problem. The owner still chases missing information. Someone still checks every answer. Nobody knows who fixes it when it breaks.
That gap is why I care about the Forward Deployed Engineer, or FDE: an engineer who works closely with a customer to understand a problem, build the right solution, and help put it to use. The job connects technical work to what the people using it actually need.
AI gives us more ways to build. It does not tell us which problem is worth solving, whether the result is correct, or who should be allowed to use it. I think the ability to make those decisions and deliver the work deserves more attention than knowing the latest model name.
What “forward deployed” means
The engineer gets close to where the work happens. That might mean spending time on-site with a customer, working alongside their team remotely, or watching how a task moves between departments. The point is to learn from the people living with the problem, rather than build from a list of features alone.
This is not a job invented for ChatGPT. Palantir described its Forward Deployed Software Engineers in 2020 as engineers who embed with customers and configure its platforms to solve their problems. Its useful distinction was one capability for many customers versus many capabilities for one customer. That is Palantir’s description of its roles, not a rule for every software team.[1]
A product engineer might improve a scheduling tool that thousands of businesses use. An FDE might work with one business to connect scheduling, customer records, and the way its staff handle exceptions. Both need engineering skill. Their starting points and responsibilities differ.
There is overlap with consulting, implementation, and solutions engineering. Titles vary by company. I would judge the work by what the person owns: understanding the problem, making technical decisions, implementing the solution, and helping the customer operate it. A sales demo or an advice deck alone does not cover that whole job.
Follow one problem past the demo
Take an event promoter who wants AI to help with sponsor inquiries. This is an illustrative example, not a customer case study.
The first request might be, “Build me an AI that answers sponsors.” An FDE should ask what is actually slowing things down. Are inquiries going unanswered? Is the sponsor information scattered? Does every proposal wait on the promoter because only they know the available placements?
Say the bottleneck is checking which sponsorship slots remain available. Generating replies faster could make things worse if the assistant promises a slot already sold. The useful first step may be cleaning up the inventory record and deciding who can change it. Existing software might already do that part.
Then AI could help turn a messy inquiry into a short brief: what the sponsor wants, which event they mean, what information is missing, and where each detail came from. The current inventory system would supply availability. The promoter would still approve prices and commitments until a narrower, tested permission was agreed.
Now the implementation has something to test. Give it an inquiry with no event date. Give it two businesses with similar names. Ask for a sold-out placement. Change the inventory after a draft is prepared. Check whether it asks, stops, or rechecks instead of making up an answer.
A working demo is only part of delivery. The staff need to know how to correct a record, pause the assistant, and find a failed task. Someone needs to own maintenance. The promoter needs to compare the new process with the old one, including review time and running costs. If the owner still untangles every inquiry, the job is not finished just because the bot replies.
Why AI makes this more relevant
The model companies themselves are investing in this work. OpenAI’s FDE role describes ownership from discovery and technical scoping through system design, build, and production rollout. Its measures include adoption and workflow impact. Anthropic’s role similarly calls for engineers to work inside customer systems and build production applications with attention to safety and reliability.[2][3]
There is a larger training effort too. In June 2026, Anthropic announced that DXC plans to train tens of thousands of Claude-certified FDEs from its existing engineering teams. That is an announced program, not a count of graduates or proof of jobs available to beginners.[4]
These are concrete examples of companies putting engineers near customers. They do not establish that FDE is the fastest-growing career or that every business needs to hire one. The reason I find them important is the work they describe: turning access to powerful models into something people can use in their existing systems.
AI can help produce a prototype, explain unfamiliar code, or draft a test. But a sponsor brief still needs the right event record. A connection to another app still needs the right permissions. A confident answer can still be wrong. Faster building gives you more chances to try an idea; it can also let you build the wrong thing faster.
For a smaller business, this creates room to test improvements that might previously have required more time or outside help. That is a possibility, not a promise of lower total cost. The saving has to survive setup, corrections, customer review, and maintenance. Sometimes the best improvement is a clearer process without AI.
The skills behind the title
I think about FDE preparation in six connected areas. These are our learning framework at cYpher.camp, not an official industry certification.
1. Business discovery and solution judgment. Can you find the problem behind the request? Can you explain when to build, use existing software, change the process, or leave the task with a person? In the sponsor example, that means finding the inventory bottleneck before building a reply bot.
2. Software and integration fundamentals. Can you understand what the code does and how systems exchange information? An API is a defined way for one system to talk to another. You need to know what happens when that connection fails, repeats a request, or loses access. AI writing the code does not remove that responsibility.
3. Data reasoning. Can you tell which record is current, what a missing value means, and whether two similar names belong to the same customer? Bad records do not become trustworthy because a model puts them into a nice paragraph.
4. AI implementation and evaluation. Can you give a model the information it needs, check its answers against representative cases, and identify when it should not answer? Evaluation means testing behavior against clear expectations, including difficult cases, rather than accepting a few impressive examples.
5. Security and operational reliability. Can you limit access, protect private information, spot failures, and recover without repeating a harmful action? Permission to draft a sponsor reply is not permission to send it, change a price, or collect payment.
6. Customer delivery and handover. Can another person use what you built, understand its limits, and take over routine operation? You need to explain the trade-offs and leave usable instructions. A system that depends on you remembering every detail is hard to hand over.
You do not need equal depth in every area before you begin learning. You do need to see the gaps. Strong communication cannot compensate for an unsafe connection, and strong coding cannot tell you whether the customer wanted the wrong thing.
A learning path, not a shortcut to a title
For a developer, FDE work may mean getting closer to customers and taking more responsibility for adoption. For someone who knows a business well, it may mean building enough technical depth to test and deliver improvements instead of only recommending them.
An owner can borrow the method without becoming an engineer. Ask a prospective builder how they will understand your process, prove the result, and leave you able to operate it. You are buying a useful improvement, not an acronym.
If you want an enterprise FDE job, read the actual requirements. The OpenAI posting I reviewed asks for five or more years of relevant experience. Anthropic’s Munich posting asks for eight or more, along with production AI and programming experience. Requirements differ, but these examples are not beginner roles.
Our direction at cYpher.camp is to teach the skills and provide usable tools through our AI Studio. I want learning to lead into work you can inspect and improve, not stop at a video or require you to assemble a separate tool stack before trying anything.
That is a learning approach, not a job guarantee. We are planning an optional assessment to help identify gaps in your judgment. It cannot certify that you can deliver a production system. Watching a course, generating code, and owning a working solution are different kinds of evidence. Practice needs to include explaining your choices and correcting a result you did not expect.
Start with a problem you can check
Choose one small recurring problem in a business you understand. Use fictional or non-sensitive inputs for the first test. For the sponsor example, write this brief before opening a tool:
Goal: turn a made-up sponsorship inquiry into a brief the promoter can review. Identify the event, requested placement, missing information, and source for each detail. Do not claim live availability.
Limits: local test only. No real sponsor data, connected accounts, sending, pricing changes, bookings, or payments. If you choose to use AI Studio, stay within your existing access and credits; drafting locally is also an option.
Check: try a complete inquiry, a missing date, a similar business name, and a request for a sold-out placement. Explain what should happen before you run each case. Record what went wrong, what you changed, and whether the original problem still needs AI.
That gives you a first piece of work and a way to judge it. A professional portfolio can build from there toward tested integrations, independent technical diagnosis, and a handover another person can use. It should show your decisions and failures, not only the polished screen.
The FDE role matters because somebody has to connect what technology can do with what a customer needs done. AI expands what we can attempt. Learn to choose the attempt, test it, and take responsibility for what comes next.