Split View: 빅테크 개발 문화 완전 가이드: Google, Meta, Amazon, Netflix, Apple은 어떻게 다른가 (2025)

Big Tech Engineering Culture Deep Dive: How Google, Meta, Amazon, Netflix, and Apple Differ (2025)

Intro — "Let's Do It Like Netflix"

Startup founder: "I want to build a Freedom-and-Responsibility culture like Netflix."

Two months later: "Why is the team a mess?"

Answer: Netflix's culture rests on the premise that only top-tier performers are on the team. The Keeper Test ("Would I fight to keep this person if they tried to leave?") only works if everyone is a Keeper. Copy the rituals without Netflix's hiring filter (senior-only, no junior hiring) and you get a disaster.

Is culture transferable? Only partially. Different org size, market, talent pool, and product maturity demand different cultures.

This post compares:

1. Five Big Tech cultures — Google, Meta, Amazon, Netflix, Apple
2. Korean Big Tech — Naver, Kakao, Coupang, Toss, Baemin
3. Core mechanisms — perf review, code review, meetings, decision-making
4. Picking the right culture for your company — by stage

Season 3 Episode 4. Last time in "Scaling Inflection Points" we talked about Conway's Law. This time, the flip side — how culture defines org structure.

Chapter 1: Google — "Data-driven + Design Doc"

1.1 Keywords

• TGIF (Thank God It's Friday): company-wide Q&A every Friday, founders answering live (scaled back after 2020).
• Design Doc: required before writing real code, 2-5 reviewers.
• Readability: an internal certification required to LGTM in a given language.
• 20% time: 20% of your week on a side project (Gmail, AdSense were born this way).

1.2 Design Doc Culture

Before a Google engineer writes important code:

1. Write a Design Doc (4-30 pages): problem statement, goals / non-goals, alternatives compared, chosen solution + rationale, expected impact (performance, cost, security).
2. Assign reviewers (usually 2-5 Staff+ engineers).
3. Iterate on feedback (1-4 weeks).
4. Approve, then code.

Pros: wrong directions caught early, organizational learning (search old docs for rationale), onboarding material for new hires.

Cons: slow (overkill for small work), writing ability disproportionately weights your engineering evaluation.

1.3 Infamy of Perf Review

Google Perf is famously complex:

• Self-assessment every half.
• Peer reviews from 3-7 colleagues.
• Calibration — managers meet to adjust grades (similar to a forced curve).
• Promotion packet — tens of pages when you go up for promo.

Stack ranking controversy: not explicit but effectively exists. Bottom-N PIP pressure.

2023 overhaul: simplified into "GRAD" (Googler Reviews and Development).

1.4 Code Review Culture

• LGTM (Looks Good To Me) = approval.
• Readability system: pass a language-specific exam to grant LGTM.
• Average of 2+ reviewers.
• Critique (internal tool): Gerrit-based, line-by-line comments.

Public Google research data: avg reviewers 1.7, avg comments 4-5, median turnaround 4 hours.

1.5 Downsides

• Slow decisions: design doc, review, approval — many layers.
• Politics: promo-driven development (only working on what ships a promo).
• Project shutdowns: Google Reader, Wave, Stadia... "Google Graveyard."

Chapter 2: Meta (Facebook) — "Move Fast and Break Things"

2.1 Keywords

• Move Fast: originally "Move fast and break things," revised in 2014 to "Move fast with stable infra."
• Bootcamp: new hires rotate through teams for 6 weeks, then pick one.
• Dogfooding: eat your own product.
• Hackathon: 48 hours, once a quarter.
• Done is better than perfect (the office poster).

2.2 The Bootcamp Oddity

Most companies: new hires are assigned to a team at hire time.

Meta: for 6 weeks you ship small projects across 3-5 teams, then pick your own team.

Pros: new hires choose with real data, they build a cross-team network, fewer post-hire matching misses.

Cons: 6 weeks of onboarding overhead, popular teams are overloaded.

2.3 Dogfooding

• Employees use Facebook, Instagram, and WhatsApp daily.
• Internal features ship to employees first.
• A healthy bug-report culture.

Product taste comes from use.

2.4 The Reality of Move Fast

Early Meta: tens of thousands of deploys per day, destructive changes shipped anyway.

Since 2014: "Move fast with stable infra." Reasons:

• Platform stability (billions of users).
• Regulation (GDPR and friends).
• Stronger review boards.

Still tens of thousands of deploys a day, but with more gating.

2.5 Perf Review

PSC (Performance Summary Cycle): half-yearly, self + peer + manager, ratings Redefines / Greatly Exceeds / Exceeds / Meets All / Meets Most / Meets Some. Stack ranking is infamous — bottom 10% lands in PIP.

2023 "Year of Efficiency": mass layoffs; the culture hardened.

Chapter 3: Amazon — "Two-Pizza + Working Backwards"

3.1 Keywords

• Leadership Principles (LP): 16 (originally 14) principles.
• Two-pizza team: small enough to be fed by two pizzas.
• Working Backwards: write the launch press release first.
• 6-page narrative memo: no PowerPoint, six pages of prose.
• OP1/OP2: twice-yearly operating plan cycle.

3.2 The 6-page Memo

Every meeting opens with 30 minutes of silent reading of a 6-page document. Then discussion.

Why no PowerPoint:

• Slides fragment thought.
• Slides center the presenter; the audience stays passive.
• Flashy visuals hide weak reasoning.

Bezos: "The narrative structure of a good memo forces better thought and better understanding."

Structure: context and problem, data and analysis, proposal and alternatives, anticipated objections and answers, appendix with raw data.

3.3 Working Backwards

For a new product:

1. Press Release dated at launch: "This product shipped today."
2. FAQ covering what customers and reporters will ask.
3. Backwards plan: what must we do now for that press release to be real?

Starts from customer value, not technical ease.

3.4 Two-Pizza Team

Bezos:

"A team should be feedable by two pizzas (about 6-10 people)."

Communication paths scale as n*(n-1)/2. Ten people = 45 paths, 20 = 190. Small teams decide faster and own more; large teams drift into politics and delay.

3.5 Leadership Principles — 16

• Customer Obsession
• Ownership
• Invent and Simplify
• Are Right, A Lot
• Learn and Be Curious
• Hire and Develop the Best
• Insist on the Highest Standards
• Think Big
• Bias for Action
• Frugality
• Earn Trust
• Dive Deep
• Have Backbone; Disagree and Commit
• Deliver Results
• Strive to be Earth's Best Employer (added)
• Success and Scale Bring Broad Responsibility (added)

Interview LP question: "Tell me about a time you showed Customer Obsession."

"Disagree and Commit": disagree, but once the call is made, execute fully. Prevents decision paralysis.

3.6 Frugality

• Desks originally made from doors (some still in use).
• No free lunch.
• Cheap flights on business trips.

The message: money goes to customer value.

Chapter 4: Netflix — "Freedom and Responsibility"

4.1 Keywords

• Culture Deck: a 125-slide 2009 deck, canonical reading in Silicon Valley.
• Keeper Test: would you fight to keep this person?
• Stunning colleagues: only top-tier teammates.
• No rules rules: unlimited vacation, no expense policy.
• Informed captain: a single named decision-maker is accountable.

4.2 No Rules Rules

Vacation: unlimited, no approval needed. Expenses: "act in Netflix's best interest." Dress code: none. Approval processes: mostly none.

Why it works: everyone is an A-player. Treat adults as adults, and they act like adults.

Why it may not: juniors need scaffolding; in hierarchical cultures the absence of rules becomes abuse.

4.3 The Keeper Test

Managers periodically ask themselves:

"If this person told me tomorrow they were leaving for another company, how hard would I fight to keep them?"

If the answer is "not very hard," it ends in termination with a generous severance.

Pro: sustained high performance. Con: psychological safety suffers; risk aversion creeps in.

4.4 Context, not Control

Managers provide context, not micromanagement. Employees decide with information. Mistakes are tolerated; repeated mistakes are not.

4.5 Informed Captain

Every decision has one Informed Captain: gather input broadly, decide solo, own the outcome. Explicitly avoids consensus. Similar in spirit to Amazon's "Disagree and Commit."

4.6 Companies That Can't Be Netflix

Reed Hastings warns:

"This culture works at the intersection of a mature company, exceptional talent, and a simple product (video streaming)."

Dangerous in early-stage startups, manufacturing, healthcare, finance.

Chapter 5: Apple — "Secrecy + Integrated Excellence"

5.1 Keywords

• Secrecy: project info gated even internally.
• DRI (Directly Responsible Individual): one name on every task.
• Functional organization: Engineering, Design, Marketing as functions, not product divisions.
• Integration: hardware + software + services designed together.
• Taste: an aesthetic bar beyond engineering correctness.

5.2 Secrecy

• Code names for projects.
• Need-to-know even across teams on the same project.
• Physical separation: restricted buildings and floors.
• NDAs: aggressive, enforced with lawsuits.

Why: competitive advantage, launch impact, IP protection.

Downsides: hard internal collaboration, staff working without context, tense press relations.

5.3 DRI Culture

Every problem, every feature, every decision carries one name:

"Who is the DRI for this button?"

That person is questioned, decides, owns the result.

Pros: no diffusion of responsibility, fast decisions, quality ownership. Cons: bottleneck, single point of failure.

5.4 Functional Organization

Most companies split into product divisions as they grow. Apple remains functional — Hardware engineering, Software engineering, Design, Marketing. iPhone is built by every function collaborating; each function is world-class at its craft; the CEO is the single integration point. Integrated product experience requires an integrated org.

5.5 Taste and Reject

Steve Jobs famously said "No" to thousands of ideas. "What we don't do defines us." It is not surprising in a review meeting to have the first design called "garbage."

Most companies build too many things. Apple focuses.

Chapter 6: Korean Big Tech — Naver, Kakao, Coupang, Toss, Baemin

6.1 Naver

• Culture: Japanese-style detail + Korean-style speed.
• Org: Cloud, Search, AI Lab as separate subsidiaries.
• Perf: attempts at absolute-scale evaluation (2020s).
• Strong: search, AI, ClovaX.
• Weak: global presence, speed of new product launches.

6.2 Kakao

• Culture: English nicknames, informal speech, "agile" is emphasized.
• Org: "Crew"-based, flat-leaning.
• Issue: the 2022 data center fire exposed infra and DR gaps, triggering culture reform demands.
• Strong: messenger ecosystem, fast experimentation.
• Weak: sprawl of subsidiaries, internal politics.

6.3 Coupang

• Culture: Amazon DNA (LP-like principles).
• Strongly data-driven: experiments, A/B, metrics.
• Working Backwards adopted.
• Strong: logistics, data, NYSE listing.
• Weak: reputation for harsh work intensity.

6.4 Toss

• Culture: Silo organization — each product team holds its own engineers, designers, and PO.
• TMT (Toss Mission Talk): quarterly company share.
• Decisions: Silo leads hold real authority.
• Strong: product polish, launch speed, UX.
• Weak: cross-silo coordination, culture maintenance under hypergrowth.

6.5 Baemin (Woowa Brothers)

• Culture: the famous "How to work well in Songpa-gu" poster.
• Specific practices for small talk, meetings, initiative.
• Org: product-centric.
• Strong: developer employer brand, engineering blog.
• Weak: cultural change since the Delivery Hero acquisition.

6.6 Common Traits

• Overtime culture improving: 52-hour workweek law has landed.
• English at work: Toss and Coupang are English-first by default.
• Engineering blogs are a branding tool.
• Junior-friendly: new-grad hiring + in-house bootcamps (Woowa Tech Course, SSAFY).
• Tenure: 2-3 years is typical; movement between Big Tech and startups is common.

Chapter 7: Perf Review Comparison

7.1 Cadence

7.2 Rating Scales

• Google: Not Enough Impact / Significant Impact / Outstanding / Transformative
• Meta: Greatly Exceeds / Exceeds / Meets All / Meets Most / Meets Some
• Amazon: Top Tier / Highly Valued / Least Effective (URA — Unregretted Attrition Rate targets)

7.3 Shadow of Stack Ranking

Microsoft killed stack ranking in 2013: intense internal competition, killed collaboration, teams moving low performers around.

Many companies still run some variant. Debate continues — necessary evil or pure evil.

Chapter 8: Code Review Comparison

8.1 Number of Reviewers

• Google: 1-5 (avg 1.7, Readability required)
• Meta: 2-3 (Phabricator)
• Amazon: 1-2 (CR required)
• Netflix: case-by-case
• Apple: not public, within functional teams

8.2 Review Style

• Google: strict; Readability; consistency beats personal taste
• Meta: pragmatic; deploy speed first
• Amazon: "Are you sure?" questions; Dive Deep
• Korea: varies widely. Toss is strict, Baemin is pragmatic.

8.3 Automation

• Google: Critique, Tricorder, hundreds of linters
• Meta: Phabricator, Sapling (VCS), Infer (static analysis)
• Amazon: CodeGuru (internal), many pipeline gates
• Typical startup: GitHub + Actions + ESLint/Prettier

Chapter 9: Meetings and Communication

9.1 Meeting Culture

• Amazon: 6-page memo + 30-min silent read + discussion — document-centric.
• Google: slides + Q&A — presentation-centric.
• Meta: short standup + Workplace posts — async-plus-short.
• Netflix: small-group meetings + "memo culture" being adopted — fewer.
• Apple: DRI-centric, small — vertical reporting.

9.2 Async vs Sync

9.3 1-on-1s

Common: weekly, 30-60 minutes.

• Google: growth coaching, quarterly OKR check-in.
• Amazon: outcome-focused, unblock work.
• Netflix: context exchange.
• Toss: frequent, in Pair PO/design/dev structure.

Chapter 10: Hiring Filters

10.1 Google

• 4-6 interview rounds (technical + leadership)
• Coding interview: algorithm-heavy
• Hiring committee owns the decision (hiring manager is not the decider)
• Levels: L3 (new grad) through L10 (VP)

10.2 Meta

• 2 coding + system design + behavioral
• Will give feedback email on request
• Levels: E3 through E9

10.3 Amazon

• "Bar Raiser": an interviewer from an outside team (bias control)
• LP-heavy behavioral
• 5-6 onsite rounds

10.4 Netflix

• Senior-only hiring (almost no early-career)
• Culture fit emphasized
• Hiring built on the Keeper Test premise
• Top-of-Market compensation

10.5 Apple

• Team-specific process, little standardization
• Informal Bar Raiser feel
• Secrecy test: no project info disclosed even in interviews

10.6 Korean Big Tech

• Naver/Kakao: coding test + 2-4 interview rounds
• Coupang: Amazon-like (LP-based)
• Toss: TMT culture interview
• Baemin: coding + personality + culture

Chapter 11: Culture Failure Cases

11.1 Uber's Toxic Culture (2017)

Problem: aggressive growth above all, internal sexual harassment cover-ups, founder behavior, Hustle culture in the extreme.

Result: founder resignation, triggered by Susan Fowler's blog post.

Lesson: culture flaws eventually surface externally.

11.2 Zappos's Holacracy Experiment

Tony Hsieh introduced Holacracy (no managers) in 2015. Result: 14% of employees quit. Not a total failure, but it did not spread.

Lesson: no one culture fits every org.

11.3 Basecamp's 2021 Politics Ban

Basecamp banned political and social topics at work. 37% of employees quit.

Lesson: culture changes destroy trust with existing employees.

Chapter 12: Culture Checklist (12 items)

• Decision speed: avg approval time on design docs (weeks)
• Meeting time share: engineer's weekly meeting hours %
• Deploy frequency: deploys per day
• Review turnaround: median PR-to-merge hours
• On-call burden: pages per week, night share
• Psychological safety: "I can speak up about mistakes" survey score
• 1-on-1 follow-through: skip rate %
• Perf transparency: criteria written down
• Career ladder: Staff/Principal definitions public
• Attrition balance: eNPS, voluntary attrition
• Async doc ratio: Slack vs Notion/Confluence
• New-hire onboarding: time to first PR

Chapter 13: 10 Culture Anti-Patterns

1) Cargo-culting

A new startup clones Netflix. Ignores the preconditions (senior-only, simple product).

2) Posters equal culture

"Be yourself" on the wall, no practice behind it. Culture is managers' daily decisions.

3) Founder-dependent culture

Culture only stands while the founder is there. Founder leaves, collapse. Culture needs to be institutionalized.

4) "We're a family"

Layoffs then feel like betrayal. Families don't fire. "Pro sports team" is honest.

5) Top performers without a bottom mechanism

Want only the best without any downside lever. Unmanageable in practice.

6) Decisions by consensus

Consensus illusion. Decisions stall. Use Disagree and Commit.

7) No anonymous surveys

Leaders stop hearing the truth. Add eNPS + anonymous surveys.

8) Perf Review is the only feedback

Silence the rest of the year. Need continuous feedback.

9) Meetings without documents

Decisions vanish; the same discussion reoccurs. Adopt design docs.

10) Hustle worship

Glorifying 72-hour weeks. Short-term works, long-term burns out the team. Sustainability wins.

Chapter 14: Which Culture Fits Your Company

14.1 Seed to Series A (5-30 people)

• Recommended: Done > Perfect, everyone reads everyone's code, informal comms.
• Avoid: mandatory design docs, tiered perf review.
• Benchmarks: early Baemin, early Stripe.

14.2 Series B to C (30-200)

• Recommended: split teams, explicit 1-on-1s, lightweight annual perf, async memos.
• Avoid: stack ranking, forcing microservices.
• Benchmarks: mid-stage Shopify, mid-stage Toss.

14.3 Mature (200-2000)

• Recommended: public level ladder, promo process, mandatory design docs, SRE culture.
• Avoid: over-centralized decisions, founder-as-bottleneck.
• Benchmarks: Notion, Figma.

14.4 Big Tech (2000+)

• Pick a base: Google-style (data + docs), Amazon-style (LPs, frugal), or Netflix-style (high-talent density — conditional on hiring filter).
• Common: explicit processes, a strong culture statement.
• Risk: bureaucracy, slower innovation.

Wrap-Up — Culture Eats Strategy for Breakfast

Peter Drucker's famous line:

"Culture eats strategy for breakfast."

No matter how good the strategy, broken culture kills execution.

Principle 1: Culture is behavior, not proclamation

Saying "we are customer obsessed" ten times is weaker than one decision to delay a feature to move customer NPS by 20% this quarter.

Principle 2: Hiring is 90% of culture

The remaining 10% is ongoing management. Hire wrong, cannot correct.

Principle 3: Culture is not transferable

Understand context; import only parts. Amazon's 6-page memo is portable. Netflix's Keeper Test is dangerous.

Principle 4: Culture evolves

Startup and enterprise need different cultures. Keep identity, change form.

Principle 5: Read the originals

• Netflix Culture Deck (2009)
• No Rules Rules — Reed Hastings
• Amazon Leadership Principles
• Working Backwards — Colin Bryar and Bill Carr
• How Google Works — Eric Schmidt, Jonathan Rosenberg
• Woowa Brothers: How to Work Well in Songpa-gu

Next Up — "From Senior to Staff, Principal, and VP: The Path to a World-Class Engineer/CTO"

Season 3 Ep 5:

• The Senior Engineer plateau
• What makes Staff+ engineers different
• A Principal Engineer's day
• IC (Individual Contributor) track vs Manager track
• The road to CTO
• Career analyses of famous engineers (Jeff Dean, Sarah Drasner, Kelsey Hightower)
• Senior engineer reality in Korea
• Job change, consulting, startup — the branching points
• Still coding at 40 and 50

See you there.