Levels, Burnout and What Comes Next

What the ladders say, and what nobody has measured

Start with what is missing, which is most of the subject. No employer, analyst or researcher has published how long FDEs stay, how many leave, why, or where they go; this book's research looked for primary attrition and tenure data and found none. Burnout evidence is anecdote, and the nearest survey covers a neighbouring role. Job posts are primary evidence of what employers ask for and T4 evidence of how the work is done; this chapter uses them mostly for the first.

The published ladders disagree about what a level is.

A ladder by years. Scale publishes three FDE levels as three separate posts. Apart from the title, the minimum years and the pay band, the three are the same text, word for word, "Technical Leadership" section included. On paper a staff engineer does a senior's work with more years behind them, and each band "may be inclusive of several career levels". That shows how one employer writes posts, not how it promotes, but it tells a candidate nothing about what changes between levels.

A ladder by behaviour. Anthropic's FDE and Staff FDE posts differ where Scale's do not. The FDE builds: "Work within customer systems to build production applications". The Staff FDE is told that "Customers bring you in to lead, not only to implement", and is asked to "Review designs and unblock other FDEs on hard technical calls" and to "Lead engagements delivered with partners, setting the architecture and the quality bar for work you don't build yourself". The requirements change from years (4+ for the FDE) to track records. The travel ask drops, from 25–50% to 25%. These are hiring asks, not promotion criteria; no employer publishes those.

No ladder in the title. OpenAI's individual FDE posts carry no level at all; only the two manager posts break the pattern. A single US band covers the title, nearly as wide as the span Scale splits into three. Palantir took this furthest. Nabeel Qureshi, who joined Palantir as an FDE in 2015 and left in 2023, describes a company that chose not to give people titles, and that as a result people "came in and out of fashion very quickly, inside the company" (Reflections on Palantir, October 2024; T4, one former employee's recollection).

The top of every ladder is narrow. In Bloomberry's count of posts that state experience, 60% ask for three to five years and only 8% for nine or more (T2; the full split is in Getting Hired). Most of the market hires the middle, which is where the volume trap in the next section catches people.

Leverage, not endurance

Put the Anthropic contrast next to the rubric from The Skill Map and any ladder becomes legible. Read that grid by who does the work:

• L2: you do it, in their world. Real data, a real sponsor, a real engagement.
• L3: you do it against someone with power. A security reviewer, a director, a sponsor who could have said no and didn't.
• L4: it happens without you. Another FDE starts from your template, the roadmap carries your pattern, the company chooses its next customer partly on your case.

That is a ladder of leverage: the higher levels are the same work done through other people, other engagements and the product, not done harder. Anthropic's Staff post says so directly: "work you don't build yourself". The manager posts in Leading FDEs push one step further.

The trap is to climb by volume instead. Volume feels like progress: a third engagement, a fourth, more customer sites, more systems you know better than anyone. But each repeat produces evidence at the level you already hold. The Skill Map warned that "a great deal of experience that feels like FDE work is L1 done many times"; the mid-career version is L2 done many times. Five engagements in which you personally built the integration, ran the eval and briefed the director are five pieces of L2 and L3 evidence. None of them is L4, because in none of them did anything happen without you.

Where a ladder is written by years, as Scale's is, or not written at all, ask your lead three questions before the next review:

• "What did the last person promoted to staff stop doing themselves?" An answer that names only tenure or customer count means the ladder rewards volume.
• "Who has started from something I built, and would they say so?" No name means no L4 evidence, however many engagements you have run.
• "Does anyone here get time between engagements, and what is it spent on?" This one also answers the load question in 18.4.

Vague answers to all three are information about the employer, not about you. Getting Hired has the versions to ask before you sign.

What the burnout evidence actually says

Burnout is the risk everybody names and nobody has measured. Here is all the evidence, graded.

One practitioner, relayed by a reporter. LeadDev's July 2026 piece on the role reports Piotr Kraus, an ex-Palantir FDE now at an FDE advisory firm, as saying the constant travel makes the role "susceptible to burnout a lot earlier than others"; the piece notes that he often worked in remote, hard-to-reach places. A recruitment-firm founder adds: "It's definitely not for everybody… It's a very challenging role" (LeadDev, July 2026; T3, and the burnout line is the reporter's paraphrase, not a direct quote). The article gives no attrition data. Its other much-quoted line, Chris Slovak of Unframe saying the model "doesn't seem very sustainable", is about the business model, scarce engineers placed in every customer, not about the people. It is not evidence of burnout.

One insider, with a counter-signal. Qureshi describes Palantir FDEs expected to "go onsite" to the customer "3-4 days per week", and says the field engineers "tend to have high pain tolerance" (T4). He also says the company "was loud about the fact that people worked long hours, were paid lower than market, and had to travel a lot", yet "it was common to stick around for 5+ years". That runs against the assumption that a hard field job must have high turnover, with a confound he names himself, the stock (Where FDEs Work reads it as a pay design). Long stays under heavy load may say more about equity than about stamina.

One adjacent survey. Consensus, which sells demo software, surveyed more than 600 presales and sales professionals and reports that "Burnout is a real and growing problem for SEs". It finds burnout higher in teams with six or more account executives, where sales engineers "struggle to keep up with growing demo demands", and at large enterprises (Consensus, September 2025; T2, vendor-reported, with no prevalence figure and no published method). Sales engineers are not FDEs. What carries across is the shape: burnout rose with the demand on each person and with repetitive work, which the report also names, not with the role as such.

One resignation that was not burnout. The disappointed hire in Gergely Orosz's reporting, who "quit 4 weeks later", left because the job was "a typical IT services mindset", not because it was exhausting (Pragmatic Engineer, March 2026; T4). Lumping that mismatch (What an FDE Is) in with burnout hides both.

That is all of it. Course sites make confident claims about weekly hours and cite nothing. The honest summary: the load is real and advertised, the damage plausible and unmeasured, and the only survey, of a neighbouring role, points at the load on each person.

The three loads, and how to cap each

What wears FDEs down comes in three loads. Employers advertise the first, barely mention the second and never mention the third.

Travel: advertised as a ceiling. Most lab posts state a travel percentage, and most of them say "up to". Bloomberry finds 68% of FDE posts require travel (T2), but puts its "builder" type at 30–50% in one place and FDEs at 50–75% in its own table, so quote neither as settled. A ceiling is not an average, and travel in an engagement is lumpy. Discovery wants you on the floor for days at a stretch (Discovery); the build can mostly run remotely; the handoff drill needs you in the room again (Production and Handoff). The book's rule of thumb, untested: cap travel by phase, not by percentage. Agree in the engagement brief which weeks are on site, and treat on-site weeks outside discovery and handoff as a signal that something else has gone wrong.

On-call tails: barely advertised. Of OpenAI's 26 forward-deployed posts on 17 October 2026, one, the financial-services post, mentions "on-call readiness". The others are silent, while the same board's infrastructure posts spell out the on-call rotation. Silence is not absence: the FDE's on-call is not a rotation anyone schedules, so nobody writes it into a post. Yet every system you build and do not hand off cleanly stays yours. After three engagements you can be answering for three systems while building a fourth. The cap is the dated support window from Production and Handoff: a scope, a channel and an end date, written down. A handoff without an end date is not a handoff. It is the second load waiting to start.

Consecutive engagements: never advertised. No post says how many engagements run back to back, or how many at once. The only public number is one team lead's remark of two to three projects per FDE (T4), quoted in Leading FDEs, which also proposes counting heavy phases rather than accounts. If the Consensus shape holds, the demand an FDE controls least is engagements with no gap between them. The cap is a gap, planned and named; the composite below asks for two to three weeks, a working figure, not a measured one. The best use of the gap is the work that produces leverage. Writing the product-feedback brief (Feeding the Product), turning a component into a template, reviewing another FDE's design: these are recovery from the field and L3–L4 evidence at the same time. An FDE who goes straight from one engagement to the next loses both.

That is why level and load belong in one plan. Climbing by volume and burning out are one mistake seen from two sides: the move that raises your level, doing more of the work through others, is the move that lowers your load.

Ask for all three caps in writing. If the answer to each is no, believe it. An employer that will not date a support window or leave a gap between engagements is probably selling your hours; the five questions in Origins and Economics will confirm it. If so, sustainability and promotion are both an argument with the business model, and the honest options are a different team or the exits below.

What comes next: the exits, and what each needs

Nobody has published where FDEs go. Bloomberry's profile sample describes where they came from, not where they went, and no employer has released alumni data. The map below is built from employers' asks and one insider's account, not outcomes. The question for each exit is which domain it draws on: an exit stays open only while you hold evidence there.

Two of these deserve a caution. The founder exit has the most enthusiastic advocates and the least data. Where FDEs Work sets out the record: one alumnus's argument, and employers, Anthropic among them ("Former technical founders are also encouraged to apply"), recruiting founders in the other direction. Nothing shows how many FDEs found companies, or how those companies fare.

The consultancy exit deserves the other caution. It is a respectable job, but it is the one you reach without choosing, one well-run engagement at a time; The Skill Map's filled self-assessment ended exactly there. Keeping exits open mostly means not letting any domain sit at L1 for a year.

One more reason to keep them open: the title may not last. A practitioner in China, quoted under a pseudonym, compares it to the product manager of the mobile-internet boom: "它大概火爆了五到十年，然后慢慢降温，我认为FDE也会经历类似的过程" ("it was hot for roughly five to ten years, then slowly cooled; I think FDE will go through something similar") (中新经纬 via 36Kr, June 2026; T4, one person's forecast relayed by a news outlet). It is one guess, but a fair reason not to let the title be your only evidence.

The template: a growth plan

The growth plan starts from the self-assessment in The Skill Map and adds what that template leaves out: the load you are carrying and the exits you are keeping. Fill it in after each engagement and take both to your lead.

GROWTH PLAN                                    date: ______________
Name / role: ______________________   Self-assessment dated: ______

1. LEVEL: CURRENT, NEXT, AND THE ENGAGEMENT THAT WOULD SHOW IT
   (current level = section 1 of the self-assessment, not hopes)
   Domain     Now  Next  Evidence that would count   Where/when
   technical  ___  ___   _________________________   __________
   customer   ___  ___   _________________________   __________
   product    ___  ___   _________________________   __________
   judgment   ___  ___   _________________________   __________

2. LEVERAGE: WHAT I STOP DOING MYSELF TO GET THERE
   1. _____________________________________ handed to: __________
   2. _____________________________________ handed to: __________

3. LOAD CHECK (last 12 months, and the next engagement)
   Weeks on site, by phase:   discovery ___  build ___  handoff ___
   On-site weeks outside discovery/handoff: ____  why: __________
   Systems I still answer for       support window ends
   ____________________________     ___________  (none = no end)
   ____________________________     ___________
   Engagements back to back with no gap: ____
   Engagements at once (incl. support tails): ____
   Gap before the next engagement: ____ weeks, used for: _________
   Red line I will raise before crossing: ________________________

4. EXITS I AM KEEPING OPEN (at least two)
   Exit             Draws on        My evidence now   Keep-alive
   ______________   _____________   _______________   ___________
   ______________   _____________   _______________   ___________
   Exit I would drift into if nothing changes: ___________________

5. WHAT I ASK MY LEAD FOR
   1. ______________________________________________________________
   2. ______________________________________________________________

6. REVIEW
   Lead agreed / changed: ________________________________________
   Lead: ______________   next review: ________

Section 2 is the one people leave blank, and the one the argument rests on: without something handed to someone else, the plan is a plan to do more. Section 3 asks for counts, not feelings, because the little evidence there is points at load, and nobody's mood in month nine is a reliable instrument.

The composite: the insurer's FDE plans the next year

The filled plan follows. Levels and one-offs are carried over unchanged from the filled self-assessment; every engagement figure is a range.

GROWTH PLAN                               date: month 9, the insurer
Name / role: the FDE, mid-level      Self-assessment dated: month 9

1. LEVEL: CURRENT, NEXT, AND THE ENGAGEMENT THAT WOULD SHOW IT
   Domain     Now  Next  Evidence that would count   Where/when
   technical  L2   L3    security review passed,     next eng.,
                         lead not in the room        weeks 1-3
   customer   L3   hold  sponsor's weekly note from  next eng.,
                         week 1; L4 needs a failing  throughout
                         engagement: do not seek one
   product    L1   L2    product-feedback brief,     gap: draft;
                         written and presented alone next eng.
                                                     by month 3
   judgment   L2   L3    a written no that holds on  next eng.
                         a 2nd engagement (first:
                         write-back, month 2)

2. LEVERAGE: WHAT I STOP DOING MYSELF TO GET THERE
   1. answering adjusters' questions   handed to: insurer ops team
      directly                         (runbook owner)
   2. fixing insurer defects by hand   handed to: ticket channel,
                                       then product team

3. LOAD CHECK (last 12 months, and the next engagement)
   Weeks on site, by phase:  discovery: most  build: 1-2 days/wk
                             handoff: drill week
   On-site weeks outside discovery/handoff: few  why: model risk
   Systems I still answer for       support window ends
   auto FNOL triage (defects)       a few months after handoff
   Engagements back to back with no gap: next one is proposed so
   Engagements at once (incl. support tails): 2 for some weeks
   Gap before the next engagement: 2-3 weeks asked, used for: the
   brief, the template write-up
   Red line I will raise before crossing: a third system on my
   name before the insurer's window closes

4. EXITS I AM KEEPING OPEN (at least two)
   Exit             Draws on        My evidence now   Keep-alive
   staff FDE        technical L4    one-off: 2nd      review one
                                    insurer reused    FDE design
                                    the template
   product role     product L3      L1 (one co-       the brief,
                                    written brief)    alone
   Exit I would drift into if nothing changes: services /
   deployment-company delivery

5. WHAT I ASK MY LEAD FOR
   1. 2-3 weeks between engagements for the brief, on the plan
   2. the next security review solo; the lead reads, not attends

6. REVIEW
   Lead agreed / changed: gap agreed at the shorter end; manager
   and founder exits not listed: no evidence yet, revisit in 12 mo
   Lead: the lead       next review: end of next engagement

Three things in the filled plan carry the chapter. Customer stays at L3 on purpose: L4 needs a failing engagement held to a decision, and nobody should plan to acquire one. The two items in section 2 are the leverage move and the load cap at once; routing the insurer's questions through its own team is what handoff was for, and it ends the on-call tail. And only exits with some evidence behind them are listed. The others are not closed; there is just nothing yet to keep them open with.

Signals you are climbing by endurance