They Just Revealed Their Internal Process for Becoming an AI FDE
🔗 Link do vídeo: https://www.youtube.com/watch?v=AD-EmZ3v6-g
🆔 ID do vídeo: AD-EmZ3v6-g
📅 Publicado em: 2026-08-07T15:54:30Z
📺 Canal: AI LABS
⏱️ Duração (ISO): PT15M4S
⏱️ Duração formatada: 00:15:04
📊 Estatísticas:
– Views: 6.758
– Likes: 247
– Comentários: 7
🏷️ Tags:
The forward deployed engineer is the AI engineer role companies can't fill fast enough. OpenAI, Anthropic and Cursor just explained how forward deployed engineering actually works, so here's the full forward deployed engineer roadmap and why the FDE is one of the fastest growing AI careers.
Meet Dex AI at https://dex.plug.dev/ecJ4eUg and get introduced to software engineering roles that actually fit.
Community with All Resources: http://ailabspro.io
The Roundup, our daily newsletter covering the AI stories that matter. Join now: https://www.theroundup.so/
Three years ago almost nobody needed a forward deployed engineer. Now postings are up 729% in a year, AWS has put a billion dollars into a whole department of them, OpenAI's own FDE team went from two people in January to thirty nine, and over a hundred Y Combinator startups are hiring for it. Colin Jarvis, who runs the forward deployed team at OpenAI, says there just aren't a lot of FDEs out there.
What is a forward deployed engineer
– Someone whose whole job is getting a business to actually use AI. Not advice, not a strategy document. Getting it running inside the systems they already have, so work done by hand gets done by AI instead.
– It started at Palantir. Their first product was software for spies, and you can't ask a spy what they do, so they showed a rough demo, got told it was terrible, wrote down every correction and built that.
– A refund agent followed the written policy perfectly and the company still lost long term customers. The person who used to do that job had an undocumented rule: anything paid on a company card got approved.
Why this has to be somebody's whole job
– MIT looked at 300 AI projects and found 95% produced no measurable return, and blamed the companies, not the models.
– Vasuman Moza, an ex Meta engineer who does this work, says AI gets slapped on top of broken processes because nobody looks at the process first. One executive burned a ten million dollar budget in three months meant to last a year.
– Palantir lost a whole year to a file format move because one engineer checked her data by double clicking the files open. Nobody knew until someone watched her work.
How OpenAI, Anthropic and Cursor run it
– Go where the volume is. OpenAI's team took the one job thousands of advisors did every day at one of the world's biggest banks, and around 98% of them ended up using it.
– Build on what the business already runs. One of Moza's clients had spent five million dollars and five years getting onto their finance system.
– Don't change how people work more than you have to, or they can't tell whether it worked and they stop using it.
– Budget more time for trust than for building. That bank build was done in six to eight weeks, then took four more months of pilots.
The forward deployed engineer roadmap, five steps
1. Watch the job being done, write down every step in order, then ask why each one exists. No concrete reason usually means a workaround nobody has questioned in years.
2. Decide which steps become AI. Fixed rules stay as ordinary software, messy judgement calls are what the model is for, and expensive mistakes stay with a person. In Moza's eight step example, four ran on their own, three had a person checking, one stayed human.
3. Build for the ways it fails, which with AI mostly means the places the model isn't sure.
4. Test it on real examples where you already know the answer, count how many it got right, then read the ones it missed.
5. Put a number on it. Cursor had someone complain an agent cost him two thousand dollars a day, until they asked what sending the wrong engineer out was costing him.
We ran those five steps on our own software company to build the internal chatbot that lets the non technical people on our team use Claude Code. Our engineers sat with HR and accounts first, and that write up turned out to be most of the build.
Hashtags:
#ai #openai #aiAgents #promptEngineering #aiEngineer #systemDesign #artificialIntelligence #palantir #forwardDeployedEngineer