Skip to content
Published on

Engineering Ladders & Compensation 2026 — Levels.fyi / Pave / Carta / Staff Engineer / Manager Path / Spotify Career Framework Deep Dive

🇰🇷KO🇺🇸EN🇯🇵JA
Split View필사 모드
Authors

1. Engineering Ladders & Comp in 2026 — The Transparency Era

Two keywords define the 2026 engineering career market: transparency and AI bifurcation. On one hand, tools and policies from Levels.fyi, Pave, and Buffer have made it easier than ever to see who gets paid what. On the other, frontier AI labs like Anthropic, OpenAI, and Google DeepMind are pushing senior-researcher packages past 1M USD, pulling the top apart from the median.

The engineering ladder is no longer a three-tier "junior, senior, manager" tree. Since Rent the Runway published its ladder in 2014, Patreon, Square, Spotify, Dropbox, and Buffer have all shared their own. The result is a shared vocabulary for what "senior at this company" actually means. On the book side, Camille Fournier's The Manager's Path and Will Larson's Staff Engineer have become de facto textbooks.

This post connects four threads: (1) compensation-market tools, (2) ladder canon, (3) AI, H-1B, and regional (Korea, Japan) trends, and (4) negotiation tactics.

2. Levels.fyi — Series A 2024, the Crowdsourced Standard

Levels.fyi started in 2017 as a side project by two engineer brothers. After going through Y Combinator in 2022 and closing a Series A in 2024, it became the de facto standard for crowdsourced compensation data. Its real strength is not just collecting salaries but maintaining a consistent level map across companies (Google L5 = Meta E5 = Amazon SDE III, for example).

Core features:

  • Per-company level, base, bonus, equity (RSU), sign-on, and total comp (TC) distributions
  • Filters by role (SWE, ML, SRE, Mobile, Data, PM, and more)
  • Regional comparisons (SF Bay, Seattle, NYC, London, Berlin, Tokyo, Seoul)
  • Levels.fyi for Teams — a SaaS that lets companies define internal levels and benchmark them against the market
  • Negotiation coaching — end-to-end negotiation consulting, reportedly priced well above 15K USD for senior packages

How to read the data:

  1. Read the distribution — median plus P75 and P90. Do not chase a single data point.
  2. Note the hire year — equity is priced at hire, so old data points have very different stock values than today's.
  3. Regional weighting — if SF is 1.0, NYC is roughly 0.95, Seattle 0.92, Austin 0.85, Seoul or Tokyo 0.6 to 0.7.

Limits are real too. Korea, Japan, and Southeast Asia still have thin samples, and the equity values reported for private startups are paper marks at the last round, not liquid cash.

3. Pave — Compensation Benchmarking with ATS Integration

Pave launched in 2020 and accelerated to a Series C by 2022. It is a B2B compensation platform that connects directly to a company's HRIS, ATS, and cap table, pulling live compensation data and benchmarking it against a market of peer companies. Where Levels.fyi data is "self-reported by employees," Pave data is "reported by companies under agreement," which gives it a very different sample-quality profile.

Key features:

  • Live market benchmarks updated weekly or monthly
  • Offer-letter generation, with equity, vesting, and FMV computed automatically
  • M&A-time compensation gap analysis for due diligence
  • Compensation cycle automation — managers move sliders for raises and the budget is allocated automatically through the workflow

Pave has become the standard tool for late-stage startups (post Series C) and pre-IPO companies. By 2026, more than 5,000 companies share data through Pave.

4. Carta — Cap Tables and Equity

Carta is fundamentally a cap-table and equity-administration SaaS rather than a "compensation tool," but because equity is the most important variable in startup compensation, any serious comp discussion ends up touching Carta.

Compensation-relevant value Carta provides:

  • 409A valuation (US IRS standard FMV), the basis for strike-price decisions
  • ISO and NSO grant administration, vesting-schedule tracking
  • Carta Total Comp (competes with Pave) — proprietary compensation benchmarks
  • An employee portal where each holder can see the current paper value of their grants

Carta had a serious credibility crisis in 2020 to 2022 over allegations that customer cap-table data was reused to run an internal brokerage. The company restructured its data governance afterwards. Even so, more than 60% of US startup cap tables sit on Carta, so it is effectively the standard.

5. OpenComp / Compa / Figures / Ravio — Other Compensation Tools

Beyond Pave, each market segment has strong specialists.

  • OpenComp — Strong at seed through Series B. Stabilized after a 2024 data-reliability rebuild.
  • Compa — Best for "negotiation data" — anonymized records of how offers actually get countered.
  • Figures — European market leader. GDPR-friendly data model and built-in conversion for 32 European currencies and tax regimes.
  • Ravio — UK-based, direct competitor to Figures. Closed a Series A in 2024, with European tech startups as the main customer base.
  • Compaas — Aimed at small teams (50 to 200 people) for compensation cycles. Slack-native.
  • CompTool — Compensation-cycle automation tied tightly to OKRs and performance data.

Selection cheatsheet:

StageRecommended
Seed to Series AOpenComp or Compaas
Series B to CPave
Series D to IPOPave plus Carta Total Comp
HQ in EuropeFigures or Ravio
Negotiation data overlayCompa

6. Glassdoor — The Classic That Survived

Glassdoor launched in 2008 and defined a generation of company review and salary sites. After acquisition by Japan's Recruit Group in 2018 it went through a quieter period, but in 2024 it strengthened anonymity policies and added AI-driven company summaries, which brought traffic back up.

Glassdoor's compensation data is self-reported, like Levels.fyi, but level mapping is weaker and equity data is patchy. By 2026 engineers typically use Glassdoor as the first quick reference and verify against Levels.fyi and Pave public reports for serious benchmarking.

On the review side it remains valuable. Manager style, work-life balance, and internal politics are signals that compensation tools never capture.

7. AngelList Stack — YC-Friendly Startup OS

AngelList began in 2010 as a matching platform for early-stage startups and investors. Between 2022 and 2024 it consolidated its products into AngelList Stack — a full-stack startup OS bundling cap-table management, SAFE issuance, fund admin, employer of record, and compensation tooling.

Engineer-facing pieces of AngelList Stack:

  • Standard SAFE templates (effectively identical to Y Combinator's)
  • Stock-plan administration — startups can issue ISO grants directly
  • Roll Up Vehicles (RUVs) — a structure that lets employees take small positions in other startups
  • Stack Talent — compensation, ATS, and offer-letter standardization

Y Combinator alignment — YC alumni companies use AngelList Stack as the near-default. Offer letters generated on Stack have remarkably consistent shape across cliff, accelerated vesting, and early-exercise options.

8. Rent the Runway Ladder — the Public Reference Standard

Surprisingly few engineering ladders are public. The most-cited reference is the one Rent the Runway open-sourced in 2014, with Camille Fournier (then CTO) as a primary author. The ideas were later expanded in The Manager's Path.

Distinguishing features:

  1. Two parallel tracks — Engineering (IC) and Engineering Management are defined as equally weighted. You do not need to become a manager to climb above senior.
  2. Three axes — Tech Skills, Getting Stuff Done, Impact, and Communication and Leadership. To move up a level you must meet expectations on every axis at that level.
  3. Level count — IC runs from L1 (Engineer I) through L7 (Distinguished Engineer); management EM1 through EM7. Most companies do not actually populate all seven and only map up to roughly L5.

A table-format ladder is a tool to give managers a shared language for borderline cases, not a literal grading rubric. It powers 1:1s, promotion committees, and interview rubrics with the same vocabulary.

9. Camille Fournier — The Manager's Path

Camille Fournier published The Manager's Path in 2017, based on her time as CTO of Rent the Runway. The book maps every step an engineer takes into management — Mentor / Tech Lead / Engineering Lead / Senior Manager / Director / VP / CTO — in a single volume.

Core messages:

  1. A Tech Lead is not a manager — Tech Leads still write code as ICs but own the team's technical direction. It is the first test on the management path.
  2. 1:1s are the manager's most important tool — frequent, agenda-driven, employee-led.
  3. Delegation versus control — the harder problem as you go senior is delegating with trust, not "I could just do it better."
  4. Promotion committees — to keep promotions fair, decisions should sit with a committee, not a single manager. That is how an org-wide bar stays consistent.
  5. The CTO's job is politics, not tech — about 80% of the role is forming alignment among the board, CEO, and other VPs.

The Manager's Path is just as valuable to ICs who want to understand what their manager is doing. Understanding the layer above changes how you negotiate.

10. Will Larson — Staff Engineer, StaffEng.com

Will Larson, after time at Stripe, Calm, and Uber as a senior engineer and engineering leader, published Staff Engineer in 2021 and the StaffEng.com site. Staff Engineer is one of the first systematic treatments of "how to climb above senior on the IC track without becoming a manager."

The four archetypes:

  1. Tech Lead — a staff engineer who owns the technical direction of one team (or a few). The most common form.
  2. Architect — designs the company's technical architecture: data models, service boundaries, infrastructure strategy.
  3. Solver — a roaming staff who lands on a single hard problem and resolves it. Requires deep trust capital.
  4. Right Hand — the right hand of a VP or CTO. Translates political and strategic work into IC-shaped output.

Why staff engineering is hard:

  • You cannot reach it with code alone — writing, design review, and consensus building are mandatory.
  • It looks like management but is not — you have no direct reports; influence is all you have.
  • It is hard to evaluate — IC metrics like lines or PR counts mean nothing at staff. The unit of evaluation is "problems we agreed to solve this quarter."

The StaffEng.com interview series is primary-source material on how actual staff and principal engineers got there.

11. Spotify Career Framework / Patreon Tracks / Square Growth

All three companies have published engineering career frameworks and they are go-to references when other companies build their own.

Spotify Career Framework

Spotify published Engineering Steps around 2018 and refined it into the current Career Framework.

Features:

  • Two tracks (IC vs. Manager) plus a matrix — IC runs Engineer 1 through 6, management EM 1 through 5
  • Five competency categories — Technical Excellence, Solving Problems, Leadership, Communication, Customer and Business Impact
  • Tied to the tribe and squad structure, so level expectations are framed as the scope of impact at squad, tribe, or company level

Patreon Engineering Tracks

Patreon open-sourced its Engineering Levels on GitHub in 2017. A clean markdown table.

Features:

  • Five IC levels (L1 Associate to L5 Principal) plus a management track
  • Each level's expectations are written across six axes: Scope, Output, Engineering excellence, Direction, Talent, and Community
  • The famous rule "you must already be doing 25% of the next level's work before promotion"

Square Growth Framework

Square (now Block) published an IC plus EM dual-track framework around 2014 to 2016. Its most distinctive feature is making lateral movement explicit.

  • Movement between IC and EM is "lateral," not "demotion"
  • Within a level, expansion of scope (team → org → company) is broken into stages
  • Written in the language of growth rather than evaluation, as the "Growth" name suggests

12. Buffer Transparency — Every Employee Salary Public

Buffer has published "every employee's compensation and the formula used to calculate it" since 2013. Crucial is not just that salaries are public but that the formula itself is published on the company site.

Components of the Buffer formula:

  1. Role base — market data by function (Glassdoor, Levels.fyi, etc.) averaged
  2. Experience multiplier — a multiplicative factor based on tenure
  3. Cost-of-living adjustment — a regional weighting (Buffer went fully remote after 2020)
  4. Loyalty multiplier — tenure at Buffer
  5. Stock options — granted separately

Effects of transparency:

  • Hiring negotiations essentially vanish — the formula determines the offer.
  • Internal equity — no friction over "what does the person next to me make."
  • Cost: the formula has to be very well designed, and changes require company-wide alignment.

37signals (Basecamp) operates a different but public formula too — "location-independent, top 10% of the market." Both companies have stayed small (under about 100 people), and the model is plausibly tied to that scale.

13. AI Engineer Comp 2026 — Frontier Labs Beyond 1M USD

The biggest 2026 shift is that frontier-AI-lab packages have decoupled from market averages.

Rough levels from public reporting and job posts:

  • Anthropic — Member of Technical Staff (MTS) at senior levels: 350K to 500K base plus equity for a total compensation of 1.0M to 1.5M USD. Senior researchers can hit 2M+.
  • OpenAI — Comparable, with PPU (profit participation unit) structures tying equity directly to company valuation.
  • Google DeepMind — Senior researchers at TC 750K to 1.2M. Lower equity share than the two above, offset by stability.
  • Meta AI — Aggressive recruiting through 2025 and 2026 for the Llama team: some senior packages cross 1M USD.
  • xAI / Mistral / Inflection — Smaller labs vary more, but senior-level packages in the 500K to 1M range are common.

Defining traits of this market:

  1. Absolute scarcity of supply — senior ML researchers, systems engineers (GPU kernels, distributed training), and alignment researchers number in the low thousands worldwide.
  2. IC roles command more than manager roles — building models directly is now valued above managing model builders.
  3. Referral bonuses at 50K to 100K USD — bringing in talent is worth so much that internal referral payouts are an order of magnitude higher than the broader market.

Big-tech senior SWEs in the US sit at TC 400K to 600K. The AI frontier adds another layer on top of that. Whether the gap closes or widens depends on GPU supply and the model-economics negotiations of late 2026.

14. US H-1B Trends — 2025 Cap Changes

The US H-1B visa runs on an annual cap of 85,000 (65,000 general plus 20,000 US-master's). USCIS introduced two big changes for 2025.

  1. Beneficiary-centric selection — even if a person applies through several companies, they get one lottery entry. The classic "apply through multiple shell consulting companies at once" path is closed.
  2. Higher fees — the I-129 fee rose from 460 to 780 USD, and added ACWIA fees pushed total company-side cost up by over 50%.

Impact across the tech industry:

  • Big tech absorbs the change with in-house legal teams. Big-tech hiring did not visibly change.
  • Smaller startups, facing higher H-1B sponsorship cost, are shifting to non-US Employer of Record models. Remote hiring through Deel, Remote, and Oyster has grown.
  • For engineers in Korea, India, and Europe, the signal is "US visas are harder." Canada's Global Skills Strategy and the UK Global Talent Visa are increasingly viable alternatives.

By 2026 the mood is "you do not have to be at HQ in the US — US big tech can hire you in Korea through EOR." That is a clearly positive shift for engineers based in Korea.

15. Korea — Toss, Kakao, LINE (LY), Coupang

By 2026 the Korean top tier on compensation roughly maps to these four company groups.

Toss (Viva Republica)

  • New grad 50 to 60M KRW base, senior (L4 to L5) 120 to 180M base plus incentive plus equity.
  • "Core track" seniors with 150 to 200M base plus sign-on plus RSU (post-IPO scenario) reach packages of 250 to 400M KRW total.
  • Negotiates aggressively; tends to match a reference offer when shown.

Kakao / Kakao Enterprise / Kakao Bank

  • Senior base 100 to 150M KRW plus equity (listed RSU). Total comp 150 to 250M KRW.
  • Subsidiary-by-subsidiary variation. Kakao Bank tilts higher on base; Kakao Enterprise tilts more equity-heavy.

LINE (LY Corporation)

  • After the 2023 LINE / Yahoo Japan merger, the Korean base is mostly LINE Plus.
  • Senior base 100 to 140M KRW plus parent-company equity. Global-traffic teams (messenger, payments) pay best.

Coupang

  • US-listed (NYSE: CPNG) so RSU is a large component. Senior base 120 to 180M KRW plus RSU.
  • Best place for "compare against Seattle / Bay Area" counteroffers; the most-frequently-reported Korea-based company on Levels.fyi.
  • Korean hiring follows US HQ leveling (L5, L6).

Below this tier sit Naver (senior 100 to 150M), Woowa Brothers (Baemin — senior 100 to 130M), and Daangn (senior 120 to 170M).

Negotiation note: Korean companies are conservative on base, but sign-on and RSU grants have real flexibility. They respond quickly to US-big-tech counters.

16. Japan — Sansan, Mercari, CyberAgent Salary Bands

Japan has historically had weak compensation transparency, but Mercari-led practices since 2020 have shifted the norm toward public salary bands.

Mercari

  • IC + Manager track matrix open since 2018, in both English and Japanese.
  • Senior SWE base 12 to 18M JPY plus RSU (publicly listed). Total comp 16 to 25M JPY.
  • Aggressive global hiring — strong representation from Korea, India, and Southeast Asia.

Sansan

  • Business-card SaaS, listed on Tokyo Stock Exchange.
  • Engineer grade system published in 2022. Five IC grades G1 to G5 plus an EM track.
  • Senior base 10 to 15M JPY plus some stock options.

CyberAgent

  • Ads, games, and media conglomerate with a strong subsidiary structure.
  • Issued an "ENG-Bond" in 2024 — a financing instrument earmarked for stronger engineer packages.
  • Senior base 12 to 18M JPY plus RSU.

Others

  • LINE (LY) — similar to Mercari; global teams pay more.
  • DeNA / Rakuten — established players, senior 10 to 15M JPY.
  • PayPay / SmartHR / freee — late-stage startups. Base slightly lower, but stock-heavy.

Yen-weakness effect: through 2024 to 2026 the weak yen has made Japanese compensation look cheaper in USD terms. To attract global talent, Japanese companies increasingly offer USD or SGD denominated packages.

17. How to Negotiate — Five Strategies

Finally, how to actually use everything above in a real negotiation.

Strategy 1: Triangulate the data

Never trust a single source. Levels.fyi (employee-reported), Pave public reports (company-reported), Glassdoor (review plus comp), and LinkedIn job-post salary ranges (mandatory in several US states) — cross-check all four to derive a P50 and P75 range.

Strategy 2: Negotiate base, bonus, and equity separately

Offers are framed in "TC," but negotiation should be itemized. Base is the most conservative dimension; equity and sign-on are far more flexible. "+30K sign-on" is much easier to win than "+10K base."

Strategy 3: Build a real BATNA or competing offer

Negotiating power comes from being able to walk away. If you do not have a competing offer, at least get your current company on the table for a retention bonus.

Strategy 4: Catch level mismatches

The largest comp jumps come from going up a level. Talk to the hiring manager openly about whether your current "L5" maps better to the new company's "L6," using Levels.fyi level maps as the shared reference.

Strategy 5: Lock in a 12-month re-look

If the initial offer is not satisfying, secure a written agreement for "performance review in 6 to 12 months with comp adjustment." Managers have more room to commit to a future adjustment, and you can use new market data to support the case later.

A closing principle: negotiation is alignment, not adversarial. The company wants you and you want the company; the number is just how that agreement is denominated. Reasoned, data-backed negotiation helps your manager too — they need data to argue your case up the chain.

18. References

Discuss on Twitter • View on GitHub