Over 150 engineers signed up for The AI FDE Onramp, a free lightning lesson we ran on Maven on September 29, 2026. The goal was to give working engineers a clear place to start on both the product side and the engineering side of the forward-deployed engineer (FDE) role, and to help them decide whether our program is right for them or their team.
If you missed it, the replay is here.
Here’s what we covered.
What We’ve Learned From Enterprise AI Work
We’ve taught AI engineering since 2024, and MLOps and ML engineering before that. In 2026 we began working much more closely with enterprises. Two engagements shaped the lesson.
A global banking firm. We ran custom cohorts that helped teams choose use cases, build them, and start shipping them to production. The lesson: the clearer you are about the problem, the more likely you are to build it successfully and get its users excited. Even the most boring problems can produce some of the biggest ROI.
A large business consulting firm. We worked with them over the summer to develop FDE-style consultants. When a small group has to solve a problem end to end, you need deep product leadership and deep engineering in the same room. Organizations can no longer get away with only one.
So What is a Forward-Deployed Engineer?
The role is confusing, and for good reason. We looked at a frontier lab’s FDE job posting, which describes owning discovery, technical scoping, system design, build, and production rollout alongside customer teams. Success is measured by production adoption, workflow impact, and eval-driven feedback.
That sounds like being good at everything. Matt Sharp (author of LLMs in Production) and Tyler Laughlin (an FDE at Adobe) helped us break it down.
- It’s multi-dimensional, and that’s not new. Strong AI engineers already had to master data, engineering, and some product management. FDEs add soft skills: finishing the project, then identifying and advocating for the next one. Many FDEs embed as consultants, and the goal is to become valuable enough that they keep you for multiple projects.
- You own the end-to-end picture. Getting from prototype to staging to thousands of users takes teams of specialists. The FDE learns how the whole stack fits together and partners with each of those teams to drive the capability through.
- Engineering is in the title for a reason. AI-assisted coding can help you build products and move quickly, but it can also leave security and architecture gaps. Understanding the architecture and knowing what you built is what separates an FDE. Matt also described loop engineering: you can take yourself out of a loop and let AI run it, but there’s always another loop, and you have to understand what happened in the last one to build the next thing.
- Think of it as three buckets. You need to be good enough at AI product management, strong at AI engineering, and equipped with domain expertise for the problem you’re solving. Companies can reduce the strain on one person by adding a product partner or subject matter experts, but the more of these hats you can wear, the further your technical expertise goes.
Matt’s perspective on “unicorn” expectations: this has happened before. Fourteen years ago it was data scientists, then data engineers, then MLEs. FDE is the latest version. The expectation is often met by several people filling each other’s gaps, and the more skills you have, the more people are willing to pay you.
A Forward-Deployed Engineer Starts By Shaping the Build
We used a map of the skills AI engineers need from DeepLearning.AI’s The Batch as a reference. It covers building and deploying AI applications, software engineering fundamentals, coding agents, and shaping the build. Rather than throw lists at you, we pointed to the last item as the place to begin.
Everyone wants agentic AI. The hard part is getting concrete about what that means and at what scale. Companies have already spent heavily on tools and tokens and are asking where the results are. This is the FDE’s opening:
- Build a small MVP for a real user with a real workflow. Pick something verifiable, even if it’s boring. AI coding tools can help you get to a quick first build.
- Get it in front of one, two, then three users. The user may not be the stakeholder who asked for it, and skipping this step is where most people fail.
- Move from zero-to-one to one-to-one-hundred. Can you attach it to a real part of the production workflow? Who is the buyer versus the user? Can you get 20 people using it daily? That evidence is what makes the case for the investment in data access, security, and governance.
- Pressure-test it. Try prompt injection, add guardrails, and get the permissions you need from the people who’ve overseen the production system for decades.
The build takes an hour. Then the product development begins.
We also shared an activity to help you get concrete about your own idea: bit.ly/FDE-concreteness. It’s the highest-leverage exercise we know that isn’t technical.
The Data Problem
The biggest concern we hear, year after year, is “I can’t get started because of the data.” We closed the lesson on this. Matt’s answer as a consultant: there’s always a way. You can’t leak PII, but you can look at redacting it before it ever becomes a worry.
This is also why our course covers local inference and synthetic data early, so you can work on real projects without sensitive data blocking progress. We’ll go deeper on this in a future post.
Meet the Team Behind Version 1.1, Cohort 11
The Onramp also introduced the people behind FDE v1.1, the 11th cohort of our AI engineering program: Dr. Greg Loughnane, Matt Sharp, Tyler Laughlin, Jacob on course operations, and Gina, who leads creative and web strategy. Subject matter experts from Salesforce and large financial institutions give explicit feedback on each project. Matt’s book, The AI Engineering Bootcamp, comes out in January 2027.
Stay Tuned
The Onramp is a starting point of the AI Forward-Deployed Engineer Certification, or AI FDE Certification. Through the AI FDE Certification you can keep getting:
- Lessons and recaps
- Cutting-edge material for the fast-moving AI landscape
- Deeper thinking on problems, product, and users
- A current enterprise technical setup to build and run production AI
Watch the replay. Or explore the certification.
Build. 🏗️ Ship. 🚢 Share. 🚀
Your Curriculum Crew at AI Makerspace
References
- The AI FDE Onramp: Replay (YouTube). https://www.youtube.com/watch?v=8Ufd1reJdIs&list=PLQy2ynNIXqkY
- FDE concreteness activity (AI Makerspace). bit.ly/FDE-concreteness
- “The AI Engineering Skills Map,” Andrew Ng, DeepLearning.AI’s The Batch. The map lays out four skill areas: building and deploying AI applications, software engineering fundamentals, using coding agents, and shaping the build. https://www.deeplearning.ai/the-batch/the-ai-engineering-skills-map deeplearning
- OpenAI Forward Deployed Engineer job posting. The Dublin posting describes owning discovery, technical scoping, system design, build, and production rollout, and measuring success through production adoption. https://openai.com/careers/forward-deployed-engineer-sydney-sydney-australia/ seedtable
- The AI Engineering Bootcamp (book), AI Makerspace. https://aimakerspace.io/the-ai-engineering-bootcamp-book/
- The AI FDE Certification (Maven course page). https://maven.com/aimakerspace/ai-fde-certification