In candidate debriefs after Amazon loops, engineers who get a "Strong No Hire" almost never fail on coding syntax. They fail because their behavioral stories stay vague, credit "the team" instead of their own work, or fall apart when the Bar Raiser asks two follow-up questions.
Amazon grades every candidate against 16 Leadership Principles (LPs). Each interviewer on your 5-person loop is assigned two or three specific principles, and their written notes decide whether you get an offer, get downleveled, or get turned down.
Here is the full playbook: all 16 Leadership Principles for software engineers, the questions interviewers actually ask, production-grade STAR examples, and how the Bar Raiser debrief works.
Complete Master List: All 16 Amazon Leadership Principles
Before your loop, map out which principles test which engineering skills:
| # | Leadership Principle | Core Engineering Focus | Primary Interview Question | Target Seniority Level |
|---|---|---|---|---|
| 1 | Customer Obsession | Prioritizing user latency and reliability over developer convenience | "Tell me about a time you had to push back on a feature request because it degraded user experience." | All Levels (L4 to L7) |
| 2 | Ownership | Acting beyond your team's direct service boundaries | "Give an example of a critical system issue you fixed that was not your team's responsibility." | All Levels (L4 to L7) |
| 3 | Invent and Simplify | Reducing operational complexity and deprecating technical debt | "Describe a time you solved a complex architectural problem by simplifying existing infrastructure." | All Levels (L4 to L7) |
| 4 | Are Right, A Lot | High-stakes technical decision-making under uncertainty | "Tell me about a time you made a major architectural bet with incomplete data. How did it play out?" | Mid to Principal (L5 to L7) |
| 5 | Learn and Be Curious | Proactive exploration of new technologies and paradigms | "Describe a scenario where you explored an unfamiliar technical framework and brought it into production." | All Levels (L4 to L7) |
| 6 | Hire and Develop the Best | Technical mentorship, code review rigor, and leveling others | "Tell me about an engineer you mentored who made a measurable jump in technical scope or leveling." | Senior to Principal (L5 to L7) |
| 7 | Insist on Highest Standards | Refusing to compromise on testing, security, and SLAs | "Tell me about a time you refused to ship a release on schedule because it failed your quality threshold." | All Levels (L4 to L7) |
| 8 | Think Big | Architecting multi-year, horizontally scalable platforms | "Describe a technical initiative you proposed that expanded scope beyond your team to an entire org." | Senior to Principal (L5 to L7) |
| 9 | Bias for Action | Calculated risk-taking during production outages | "Tell me about a time you had to make a critical production fix without waiting for manager approval." | All Levels (L4 to L7) |
| 10 | Frugality | Optimizing cloud compute, storage, and operational costs | "Describe a time you delivered a high-throughput service while drastically reducing AWS cloud spend." | All Levels (L4 to L7) |
| 11 | Earn Trust | Transparency during post-mortems and admitting mistakes | "Tell me about a time you made a significant technical mistake that affected production. How did you handle it?" | All Levels (L4 to L7) |
| 12 | Dive Deep | Root-cause analysis down to kernel logs and query plans | "Give an example of a persistent bug that other engineers gave up on. How did you isolate the root cause?" | All Levels (L4 to L7) |
| 13 | Have Backbone; Disagree & Commit | Standing your ground with data, then executing flawlessly | "Describe a time you strongly disagreed with your Tech Lead or Manager on architecture. What was the outcome?" | All Levels (L4 to L7) |
| 14 | Deliver Results | Hitting mission-critical deadlines under tight constraints | "Tell me about a time an external vendor failure threatened your release deadline. How did you hit the date?" | All Levels (L4 to L7) |
| 15 | Earth's Best Employer (2021) | Eliminating on-call burnout, runbook health, engineering equity | "How have you improved on-call rotations, runbook automation, or team sustainability for your peers?" | Senior to Principal (L5 to L7) |
| 16 | Success & Scale Broad Resp. (2021) | Second-order blast radius, customer privacy, global resilience | "Describe an architectural choice you made where you accounted for long-term downstream societal or privacy risks." | Staff to Principal (L6 to L7) |
Behind Closed Doors: How the Bar Raiser Debrief Works
An Amazon debrief is not a casual vote. It is an evidence-based review session run against strict internal rubrics.
1. The Scorecard Rubric and 48-Hour Feedback Lock
Each interviewer logs detailed behavioral notes and an independent rating (Strong Hire, Hire, Inclined Not to Hire, or Strong No Hire) before they can see what anyone else wrote. If one interviewer gives a Hire on coding but a Strong No Hire on behavioral evidence, that gap becomes the focal point of the debrief.
2. The Bar Raiser's True Role
The Bar Raiser comes from an unrelated division (for example, an AWS database engineer interviewing a Prime Video candidate) and has no urgency to fill the seat. Their job is to answer one question: Is this candidate stronger than at least 50% of current Amazon engineers at this level?
Amazon Interview Response Time: 2-and-5 Rule & Loop Results
Because the Bar Raiser holds veto power, a single weak pattern on Ownership, Dive Deep, or Earn Trust can block an offer even when the hiring manager wants to hire you. Once the debrief wraps, recruiting follows strict turnaround targets. Check our benchmark on Amazon interview response times and the 2-and-5 day rule to track your timeline.
3. Leveling Decisions (L4 vs. L5 vs. L6 vs. L7)
Once you clear the hiring bar, the panel decides your level based on the scope of your STAR stories:
| Amazon Level | Scope Evaluated in Debrief |
|---|---|
| L4 (SDE I/II) | Clean tactical execution, unit testing, and delivering well-scoped tickets |
| L5 (SDE II/Senior) | End-to-end service design, cross-service dependencies, mentoring juniors, and production on-call |
| L6 (Senior/Lead) | Multi-team architecture standards, org-wide best practices, and coaching engineers to promotion |
| L7 (Principal Engineer) | Multi-year technical strategy, company-wide resiliency, and resolving cross-VP design disputes |
If you interview for L5 or L6 but your stories only show local bug fixes, the panel will downlevel your offer. Review our guide on how to push back against L6 to L5 downleveling and check Amazon software engineer salary bands to see what each level pays.
The 10-60-20 STAR Formula for Software Engineers
Generic career blogs tell you to split STAR into four equal 25% chunks (see our complete STAR interview framework for the baseline). In an Amazon loop, spending two minutes on company background will get you cut off fast. Use the 10-60-20 split instead:
| STAR Component | Target Duration | Evaluator Objective | What They Score in Debrief |
|---|---|---|---|
| Situation | 10% (25 to 30 seconds) | Context clarity and scope | Technical baseline: throughput, service name, and scale |
| Task | 10% (25 to 30 seconds) | Individual accountability | Exact ownership: what were YOU responsible for solving? |
| Action | 60% (2.5 to 3 minutes) | Technical judgment and execution | Algorithms chosen, architecture trade-offs, and design debates |
| Result | 20% (45 to 60 seconds) | Quantified business impact | Hard numbers: p99 latency drop, cost saved, or SLA uptime |
The "I vs. We" Rule
Amazon hires individual builders, not vague team efforts. If you say "We migrated our cache to Redis and re-indexed Postgres," the interviewer writes down that your personal role is unclear.
Say what you personally did: "I ran the profiling trace, isolated the lock contention in Postgres, and wrote the migration script to move sessions into Redis."
All 16 Leadership Principles: Real Engineering STAR Scripts
Use these 16 engineering examples to model the depth, personal ownership, and hard numbers Bar Raisers look for.
Amazon SDE Salary 2026: SDE 1, SDE 2 & SDE 3 Pay + 5/15/40/40
1. Customer Obsession
Skip generic claims about "caring about users." Show a time you gave up engineering convenience or pushed back on a deadline to protect the user experience.
- Situation: During peak Q4 traffic, our checkout service hit a 3.2% error rate from timeouts in a third-party tax API.
- Task: As primary on-call, I had to restore reliable checkout without trapping shoppers in failed payment loops.
- Action: I rejected the quick fix of raising HTTP client timeouts, which would have tied up worker threads and crashed the gateway. Instead, I built an async fallback using a regional tax-estimate lookup in Redis. That let checkouts finish in under 80ms while a background job settled exact tax totals within 10 minutes.
- Result: Checkout errors fell from 3.2% to 0.02%, saving over $420,000 in at-risk cart volume during a 48-hour vendor outage.
2. Ownership
Do not pick a routine Jira ticket assigned to you. Ownership means stepping across team boundaries without being asked because production needed a fix.
- Situation: While debugging latency in our auth service, I saw an upstream identity service run by another division firing 14 redundant database queries per token check.
- Task: That service sat outside my team, and their backlog was frozen for six months.
- Action: Instead of filing a ticket that would sit untouched, I cloned their repo and profiled query plans in staging. I found an N+1 query bug in their ORM layer, wrote a batch query refactor with integration tests, and walked their principal engineer through the diff in a 30-minute review.
- Result: Their database CPU load dropped 44%, and our end-to-end token latency fell by 65ms across 12 million daily logins.
3. Invent and Simplify
Resist the urge to brag about a 10-microservice architecture. Amazon rewards engineers who strip out moving parts and delete unnecessary code.
- Situation: Our ETL pipeline had grown to 11 microservices and separate brokers, burning 15 hours of manual on-call triage every week.
- Task: I needed to redesign ingestion for 3x traffic growth without expanding our infrastructure footprint.
- Action: I mapped the data flow and proved 8 of those services were stateless row mappings. Rather than adding a new orchestration layer, I merged all 8 into one modular containerized worker and replaced three Kafka topics with an in-memory ring buffer.
- Result: We deleted over 14,000 lines of boilerplate code, cut pipeline latency from 18 seconds to 1.4 seconds, and saved $115,000 a year in AWS compute.
4. Are Right, A Lot
Never claim you are always right. Show first-principles technical judgment when consensus pushes the wrong way, backed by data.
- Situation: Our org planned to migrate our main transactional store from Postgres to a NoSQL document database because leadership assumed writes had outgrown relational limits.
- Task: I was the only senior engineer arguing that NoSQL would introduce data consistency bugs in our multi-currency accounting ledger.
- Action: I spent three days benchmarking our real write profile and proved 80% of write locks came from index bloat and un-sharded audit logs, not Postgres itself. I then built a working Citus table-partitioning prototype showing 10x write headroom while keeping strict ACID guarantees.
- Result: Our VP halted the NoSQL migration, saving roughly 9 months of engineering work. Two years later, partitioned Postgres handles 4x that peak load with zero ledger mismatches.
5. Learn and Be Curious
Talking about an online course falls flat. Show self-directed technical research that you put into production to solve a real bottleneck.
- Situation: Our mobile backend suffered from battery drain and bloated REST JSON payloads over weak cellular networks.
- Task: Nobody on the team had production experience with binary serialization or gRPC over HTTP/2.
- Action: Over two weeks, I benchmarked Protocol Buffers, FlatBuffers, and JSON under simulated packet loss. I wrote up latency, CPU overhead, and bandwidth numbers, then ran a live demo for our platform team showing a 70% drop in payload transit time.
- Result: I led the staged Protobuf rollout across our mobile gateway, shrinking average API payloads by 64% and cutting network errors in low-connectivity regions by 28%.
6. Hire and Develop the Best
Saying you answered an intern's questions is too small. Show how you systematically raised the engineering bar for people around you.
- Situation: Two mid-level engineers on my team kept having design docs rejected because they missed distributed system failure modes.
- Task: I wanted to build their core architecture skills instead of rewriting their docs myself.
- Action: I started a weekly "Failure Mode Architecture" workshop where we dissected past post-mortems, split-brain bugs, backpressure, and circuit breakers. I also wrote a team Design Doc Checklist covering observability, rollbacks, and safe data migrations.
- Result: Within 6 months, both engineers passed org-wide principal review on their first try, and both earned promotions to Senior SDE (L5) within a year.
7. Insist on the Highest Standards
This is not about being a perfectionist who misses dates. It means holding the line on hidden production risk even under launch pressure.
- Situation: Two days before a major enterprise launch, integration tests passed, but load test logs showed a 0.05% race condition in inventory reservations.
- Task: Product leadership asked me to sign off on launch and patch the race condition afterward.
- Action: I refused to approve the release and showed the math: at launch-day concurrency, 0.05% meant hundreds of oversold physical items. I put together a fix plan, implemented distributed Redis locks with optimistic concurrency controls overnight, and re-ran the 10,000-user load suite.
- Result: We delayed launch by 18 hours and processed 180,000 orders in the first 24 hours with zero inventory errors.
8. Think Big
Small local tweaks do not pass Think Big. Show a multi-year architecture shift that changed how multiple teams operate.
- Situation: Our video transcoding team spent $60,000 a month on on-demand EC2 GPU instances sitting at 15% average utilization between batch jobs.
- Task: My original ticket was just to write a cron script to shut instances down at night.
- Action: I stepped back and designed a serverless transcoding platform on AWS Fargate with spot compute and event-driven SQS scaling. I then partnered with three other media divisions running their own isolated GPU farms and pitched one shared multi-tenant platform to our VP.
- Result: Four teams moved onto the unified platform, cutting corporate compute spend by $1.8 million a year while raising peak transcoding throughput by 400%.
9. Bias for Action
Amazon separates reversible "Two-Way Doors" from irreversible "One-Way Doors." Show fast, calculated execution on a reversible decision.
- Situation: On a Friday afternoon, our search indexing pipeline stalled on an unhandled product metadata schema change, blocking new inventory from the storefront.
- Task: The lead who wrote the parser was on vacation, and normal process meant waiting for an emergency triage call.
- Action: Because a deployment rollback was a reversible Two-Way Door, I verified it would restore 98% of valid inventory while routing the broken 2% to a Dead Letter Queue (DLQ). I rolled back within 8 minutes, added DLQ alerts, and wrote a hotfix to sanitize the new schema.
- Result: Storefront inventory recovered in 10 minutes, protecting roughly $80,000 in sales, and the permanent fix shipped cleanly on Monday morning.
10. Frugality
Frugality is never about skipping security or tests. It means getting more output out of less infrastructure spend.
- Situation: As traffic grew to 50 million daily active users, our third-party APM and log ingestion bill jumped past $85,000 a month.
- Task: Leadership asked us to cut monitoring spend by 40% without losing outage visibility.
- Action: I audited our observability traffic and found 75% of ingested logs were routine HTTP 200 health checks and noisy batch debug traces. I set up tail-based sampling that kept 100% of error traces (4xx/5xx) and slow requests while sampling healthy traffic down to 1%.
- Result: Total log volume dropped 68%, cutting our monthly APM bill from $85,000 to $26,000 ($700,000+ saved per year) while keeping mean time to isolate incidents under 4 minutes.
11. Earn Trust
Skip hero stories where you did everything right. Earn Trust shines when you own a real mistake transparently and fix the system behind it.
- Situation: I approved and merged a pull request with an unindexed migration that locked our user database table for 14 minutes during peak European hours.
- Task: I needed to own the outage, fix the root gap, and rebuild trust with our infrastructure partners.
- Action: In our Correction of Error review, I took full accountability instead of blaming the PR author or CI tests. I documented how my review missed production lock simulation, then added automated DDL lock checks to CI/CD to block non-concurrent index changes.
- Result: Our Director of Engineering held up the post-mortem as a model of blameless accountability, and the CI check blocked three similar table-lock migrations across the department over the next six months.
12. Dive Deep
Stopping at "CPU was high" will not clear Dive Deep. Show how you dug down to kernel parameters, thread dumps, or query plans.
- Situation: Our distributed cache cluster hit 200ms latency spikes every 45 minutes while CPU and memory stayed under 30%.
- Task: Two senior engineers wrote it off as cloud network jitter, and I refused to stop there.
- Action: I attached an eBPF profiler to production nodes to inspect kernel syscall timings and socket queues. I traced the stalls to Linux Transparent Huge Pages running background memory defragmentation sweeps that paused threads for up to 180ms.
- Result: I disabled Transparent Huge Pages on our cache fleet and tuned virtual memory compaction. The 200ms spikes vanished completely, locking p99 latency at 1.8ms.
13. Have Backbone; Disagree and Commit
Do not fold early, and do not drag your feet after losing an argument. Show data-backed pushback followed by full execution once leadership decides.
- Situation: My manager and product lead wanted to adopt a closed-source workflow engine to hit a tight quarterly launch date.
- Task: I saw serious lock-in risk and steep per-execution licensing fees once volume scaled.
- Action: I spent three days writing a comparison doc covering lock-in, 5x cost curves, and an open-source Temporal alternative, and argued my case across two design reviews. Leadership still chose the vendor to hit the date. I committed 100%, owned the SDK integration, built circuit breakers, and shipped a week early.
- Result: Launch hit the deadline on time. Six months later, when vendor costs spiked as predicted, leadership used my original doc to migrate cleanly to Temporal.
14. Deliver Results
Working long hours on a failed project does not score here. Show how you cleared an unexpected blocker to hit a hard business deadline.
- Situation: Three weeks before our SOC 2 compliance audit, a zero-day hit our primary encryption library across 24 microservices.
- Task: Delaying the audit would have put several multi-million-dollar enterprise deals at risk.
- Action: Manual updates across 24 repos would have taken six weeks, so I wrote an automated AST codemod script to refactor the deprecated crypto calls, update imports, and bump versions across every repo. I then coordinated a 48-hour staging verification push.
- Result: All 24 services shipped to production four days before the audit, and we cleared SOC 2 with zero exceptions.
15. Strive to Be Earth's Best Employer (Added 2021)
This is not HR fluff. For engineers, it means cutting alert fatigue, fixing broken runbooks, and making on-call sustainable.
- Situation: On-call engineers on our team averaged 42 non-actionable pages a week, driving burnout and attrition.
- Task: I set out to strip away operational toil and fix our rotation health.
- Action: I audited 60 days of PagerDuty logs and grouped alerts by root cause. I retired 18 noisy threshold alerts, routed 14 informational notices into batched Slack digests, and wrote self-healing scripts for the 6 most common disk and cache issues.
- Result: Weekly pages fell from 42 to 3 per engineer, and voluntary attrition on the team dropped to 0% over the next 18 months.
16. Success and Scale Bring Broad Responsibility (Added 2021)
At Amazon's scale, technical choices carry real privacy, security, and downstream ecosystem risk. Show that you think past short-term speed.
- Situation: Our data team designed a clickstream pipeline to log detailed user interaction data across our global marketplace.
- Task: As lead architect on the review committee, I needed to weigh the downstream privacy risks of that design.
- Action: I flagged that storing raw IP addresses, user agents, and exact GPS coordinates was technically legal under our terms but created a severe breach risk. I wrote a new standard requiring client-side IP truncation, differential privacy noise on location data, and 30-day automated cryptographic deletion of raw session metadata.
- Result: The consumer division adopted the standard company-wide. When regional privacy laws tightened a year later, our platforms already complied with zero emergency rewrites.
How to Handle Bar Raiser Follow-Up Probes
When an Amazon interviewer says "Tell me more about that," they are testing whether your story is real or memorized. Expect three layers of follow-up questions:
| Probe Level | Bar Raiser Objective | Typical Follow-Up Question | Counter-Strategy |
|---|---|---|---|
| Level 1: Tactical | Verify authentic personal contribution | "What code did YOU write versus observe?" | Name exact PR diffs, services deployed, and code owned |
| Level 2: Analytical | Test architectural decision-making | "Why that approach over alternative X?" | Cite rejected options, latency numbers, and trade-offs |
| Level 3: Reflective | Assess self-awareness and learning | "What broke, and what would you change today?" | Share real post-mortem lessons and architectural hindsight |
Level 1: Defending Individual Contribution
"To be specific: my two teammates handled the frontend UI, while I owned the backend distributed locking layer. I wrote the Redis Lua scripts for atomicity, built the unit tests, and ran the production canary rollout."
Level 2: Explaining Architectural Trade-Offs
"We compared Apache Kafka and AWS SQS for this pipeline. We picked SQS because peak throughput stayed under 5,000 messages per second, and running Kafka clusters would have added heavy maintenance overhead for a three-person team."
Level 3: Reflecting on Mistakes and Hindsight
"If I built this system again today, I would skip JSON over the wire. At peak load, JSON parsing ate 22% of our server CPU. Starting with Protocol Buffers from day one would have saved us a six-month migration later."
