The Interview Spine · Chapter 11 of 13
What the loops actually look like, company by company
Only a handful of employers publish their process. Here is what they say, what is merely reported, and why the difference decides what you prepare for.
Most interview advice describes a generic loop that belongs to no company. A small number of employers publish their actual process — round counts, durations, what each round assesses — and that published material is both more useful and more reliable than any reconstruction. This chapter is what they say, with the reported-but-unconfirmed clearly separated.
The distinction that decides what you prepare for: published versus reported.
Published means it is on the company's own careers or engineering page and you can go and read it. It is what you will actually be held to.
Reported means a journalist saw an internal memo, or a prep vendor aggregated candidate accounts. Often accurate, sometimes narrow, and occasionally in direct conflict with the same company's own page.
That conflict is not hypothetical — there is a live one right now, described later in this chapter, where widely-syndicated reporting says one thing and the employer's live candidate page says the opposite.
So: read the company's own page, note the date, and ask your recruiter to confirm anything that matters. That habit is worth more than any amount of memorised loop structure.
The most completely documented loop
Amazon publishes role- and level-specific preparation pages on amazon.jobs, which makes it the best available window into how a bar moves with seniority — and it is worth reading directly rather than through a prep vendor:
| Track | Pre-loop gate | The loop |
|---|---|---|
| Mid-level engineer | An automated online assessment — roughly 90 minutes of coding, then ~20 minutes of system-design scenarios, then a short work-style survey tied to the company's leadership principles | Four × 55 minutes |
| Senior engineer | A 60-minute phone screen with a senior leader — explicitly half behavioural, half coding and system design | Five × 55 minutes |
| Engineering manager | The same 60-minute screen, plus a writing assessment sent two days before the loop | Five × 55 minutes |
| Security engineer | One or two 60-minute technical screens | Five × 60 minutes |
| Applied scientist | One or two 60-minute technical screens | Four × 55 minutes, including a technical talk |
Read the pattern rather than the numbers. Mid-level is gated by a machine and assessed on producing correct systems. Senior is gated by a human senior leader and assessed on judgement and trade-offs. Manager adds a written artefact. That progression — automated filter, then human judgement, then writing — recurs across employers even where the round counts differ.
Also published: system-design assessment objectives named as practicality, accuracy, efficiency, reliability, optimisation and scalability, split into high-level and low-level design; a decision within five business days; and sixteen leadership principles — not fourteen. Two were added in mid-2021 and a great deal of preparation material still teaches the old list.
The bar raiser, and why it is a structural mechanism rather than a person
Amazon publishes how the role works, and the mechanics are more interesting than the mystique.
The stated goal: every new hire should have growth potential and bring skills better than half of their would-be peers in similar roles. That is a bar defined relative to the existing team, which is why it can rise over time.
They are drawn from outside the hiring team, nominated by managers and peers, and train by shadowing for anywhere from three months to a year. There are more than three thousand six hundred of them.
And the load-bearing detail: they are a decision driver, not an adviser. The final hiring decision is made jointly with the hiring manager — which is, in effect, a structural veto on a hiring manager's enthusiasm under headcount pressure.
Two other employers achieve the same thing differently, with a hiring committee that operates independently of the people who did the interviewing. Whatever it is called, the mechanism is the same: someone with no stake in filling this role has to agree.
What other employers publish
- Google: six to eight weeks end to end; a cap of three applications per rolling thirty days; a hiring assessment plus role-related assessments; sometimes project work — a case study, writing or code samples — before the interviews; and a panel. It states that every candidate is assessed against the same published rubric. It also says, twice and unambiguously, that it does not ask brain teasers and has not for years — because its own data showed they did not predict job performance.
- Meta: a 45-minute screen structured as five minutes of introduction, thirty-five minutes on two coding problems, five minutes for questions; then a loop of three 45-minute rounds — coding, design, behavioural. Its design round is matched to your background and almost never involves writing code.
- Databricks publishes the clearest round taxonomy anywhere: separate Coding, Algorithm, System Programming, Architecture, Domain Deep Dive and cross-functional rounds, in a panel of four to six hours. Its system-programming round is assessed in pseudocode rather than a specific language, and its database track includes a seminar-style open-book round on published papers.
- OpenAI: four to six hours of final interviews with four to six people over one to two days; virtual by default with onsite as the candidate's choice; and assessment formats that vary by team — pair coding, take-home projects, technical tests.
- Shopify opens with a one-hour conversational ‘life story’ round scored on impact, readiness, trust, engagement and self-awareness, then a 40-minute coding exercise, a 75–90 minute pair-programming session with an explicit driver-and-observer model, and a technical deep dive. Candidates may use documentation and search — but not look up an existing solution.
- Atlassian: a two-part coding round (data structures, then code design), a 60-minute systems design round, a manager round, and a values round deliberately run by someone from a different department to the hiring team — with an independent hiring committee at the end.
And a large number of very well-known employers publish nothing at all about their process — including several household names in payments, cloud, streaming and hardware. For those, and for most capability-centre roles in India, the loop has to be reconstructed from candidate accounts. That is legitimate; presenting it as company policy is not.
Check yourself