Skip to content

FDE is not a role

What is the most valuable thing I could be working on?

I spent seven years at Palantir, and for nearly all of them my title was FDE. Under that one title, I did SRE work and led deployments as a tech lead. I managed a large team. I wrote backend code on a product team. I did pre-sales. I also did something close to product management across a bunch of deployments. I didn't do these all at once. I did them one after another, as the answer to one question changed.

What is the most valuable thing I could be working on?

I've come back to that question a lot this year, with founders, with engineers and with new FDEs. I now think it is the job.

In The fast layer and the slow layer, I split what makes a good FDE into two layers. Technical skill changes fast. The craft of impl barely changes. At Palantir, the broad title let each of us find where we could add the most value, and go do it. Asking the question belongs to the slow layer, whereas the answer belongs to the fast one.

Two time scales

I advise this startup, where the founders and one of their first engineers had been trying to pin down her role. They had tried a split of her time and a list of responsibilities. Neither had settled it.

There are two timescales at play here. One is how often the most valuable thing to work on changes. The other is how often you can change a role definition. At a startup, the first is much shorter than the second.

There's a human side too. Once someone hands you a role, you stop asking what you should be doing. That's natural, because the question has been answered for you. In a mature company that's fine, since the answer moves slowly. In a young one, people end up doing last quarter's most valuable thing.

So I suggested they stop defining the role. They could trust her to keep asking what the most valuable thing was. If she and the founders disagreed on the answer, that was the conversation to have.

A company still has to decide what it wants FDEs for. That was the point of "We need FDEs" is not a complete sentence. The purpose comes from the company. Inside it, the person in the seat keeps answering the question.

Valuable, and yours

The question has two halves. The first is about the outcome. What matters most right now? The second is about you. Of the things that matter, which need you in particular? I sometimes put it as: what is the most valuable thing I could do within my constraints? The constraints are your skills, the people around you, your time and what gives you energy, keeps you driven.

If you only ask what's valuable, you try to do everything, or you do work that would have happened anyway. If you only ask what you're good at, you keep doing it after it has stopped mattering.

My own first step into the FDE role came from the second half. In my first week at Palantir, I sat next to an FDE who was debugging a pipeline written in Python. He didn't know Python, and I had some experience from grad school. I wasn't the best engineer in the building. I was the best placed person at that desk. The next day, I was on a flight to Zurich to work with the client's data scientist.

Palantir had language for this. People talked about your superpower and your kryptonite. Ideally, you worked only on your superpowers, and you stayed away from your kryptonite. That fitted how Palantir hired. It looked for spiky people, who were exceptional at a few things. It was almost their job to work out where they could be most valuable, and go there. The company trusted them to do it.

What moves the answer

The question stays the same. The answer moves, and I see four things move it.

The company changes. When I joined Palantir's London office, it had about a hundred people, and almost all of them were FDEs. The platform was young, and being with customers was how most of us added value. As Foundry matured, many FDEs moved into software engineering, because that was now where they had the most leverage. By the time I left in 2022, the FDE role had become boring (to me), because the platform actually worked. Sometimes the honest answer to the question is to stop being an FDE.

The team changes. On one of my deployments there were four of us. I wrote about the pair at its centre in Every deployment is a startup with two founders. Each of us found the place where we added the most value. One strategist wrote much of the PySpark and was the face of the deployment. The other wasn't technical, but he opened pull requests to force things along. The other engineer took focused pieces of work, like making slow code fast. I handled the technical conversations with the client and pushed product fixes through. That was the most leveraged use of my time. Put four different people on that deployment, and the places would have been different. When someone strong joins a team, the answer shifts for everyone around them.

You change. Mid-career at Palantir, I managed a large team. It was impactful work. It was also painful, and I burnt out. I moved to a product team and went back to writing backend code. So the question has a second time scale. What's most valuable this week matters. So does how you keep your impact going over years. I know my own failure mode here. I go after the most valuable thing without asking whether I'll be happy doing it, and then I burn out. Your preferences will change, and that's fine. They still belong in the answer.

The tools change. In my physics work this year, the line between what I do and what Claude does has moved three times. What's left for me is choosing the problem, designing the checks and deciding whether a result is interesting. As I wrote in Authority to surprise us, deciding what to test took as long as ever. That's where my costliest mistakes sat. For FDEs, agents are now part of the team around you. Each time the models improve, the answer to "what only I can do" moves. When building gets cheap, choosing becomes the job.

How to ask it

A few habits help.

  • Ask it often. At a startup, the answer can change from one week to the next. Don't wait for a crunch, or a review.
  • Ask what happens if you don't. If the work would happen anyway, someone else is better placed to do it. Look for the valuable thing where the team says, "if only there was someone else who could build it".
  • Think about time scales. Some work runs for six months. Some can wait a week. Know which is which before you drop anything.
  • Say what you're dropping. The question is as much about what to stop as what to start. Saying no to valuable work is part of the answer.
  • Say your answer out loud. Tell your partner and your lead. If they disagree, you've found the conversation worth having.

What it needs from the company

The question only works if people have room to ask it. Palantir gave us a broad title and the trust to use it. It also gave us headroom. On that four-person deployment, I had one deployment. With two more on top, I'd have had too much in my head to work out what actually mattered. That's one reason many-to-many staffing goes wrong.

You can also hire for it. In my 2024 hiring post, the first essential was being outcome oriented. One of the unforgivables was waiting to be told what to do.

Asking it of myself

I'm still working this out for myself. I've been asking the question for more than a decade, and right now I find it as hard as ever.

I'm weighing some very different answers. I could do hands-on, part-time work at a startup. I could become an operator for a VC firm, working with the founders it backs. I could spend more of my time on physics, which I came back to last year after a decade away. Or I could go back to a full-time role.

All of them would be valuable. The hard part is the second half of the question. Which of them needs me in particular? Which would give me energy, rather than drain it? I know what I want from my work. I want control of my time while my kids are young, hard problems that need real thought, and people I like and respect. Each option offers a different mix of those, and I haven't settled on one.

In 2024, I wrote a post about my own career that ended without a conclusion. Two years on, I still don't have one.

It doesn't get easier. Keep asking.