The Onsite Loop for Mid-Level
Typical loop composition at L4/E4/SDE2 — coding, design (company-dependent), behavioral — plus how debriefs, hiring committees, and team match actually work.
What a mid-level loop looks like
The onsite loop is 4–6 rounds of 45–60 minutes each, in one day, virtual or (increasingly again) in person. A typical mid-level composition:
- 2–3 coding rounds — medium-to-hard data-structures-and-algorithms problems. Note that Big Tech made these somewhat harder in response to AI cheating.
- 0–1 design round — system or low-level design, depending on company (details below).
- 1 behavioral round — stories about your past work, scored against the company's values.
- 1 AI-assisted coding round at some companies — new enough that it gets its own lesson next.
The big three, compared
Loop composition is company-specific, and knowing the differences changes your prep:
| Google (L4) | Meta (E4) | Amazon (SDE2) | |
|---|---|---|---|
| Coding rounds | 2–3 | 2 (AI round replacing one in the 2025–2026 rollout) | 2 |
| System design | None — reserved for L5+ | 1 ("Pirate" round) | 1 system-design-ish + 1 logical/low-level design |
| Behavioral | 1 "Googleyness & Leadership" round | 1 "Jedi" round | Leadership Principles questions woven into every round |
| Special mechanic | Domain round sometimes replaces a coding round | AI-enabled round in most backend/ops loops from 2026 | A "Bar Raiser" interviewer from outside the team, with veto power |
Three practical consequences:
- Interviewing at Google L4? Do not sink weeks into system design — there is no dedicated round. Spend that time on coding depth, which is also what the hiring committee weights most.
- Interviewing at Amazon? Behavioral prep is not one round's worth of work. Every interviewer is assigned 2–3 Leadership Principles and will ask stories against them, with follow-ups. Your story bank (built in the behavioral module) gets used five times in one day.
- Always verify the actual loop with your recruiter — including whether any round permits AI. Loops change, pilots expand, and the recruiter will tell you exactly what you are walking into if you ask.
What "design round at mid-level" means
Where a design round exists (Meta, Amazon, most non-Google companies), the mid-level bar is scoped component design, not org-level architecture. Think: design a URL shortener, a rate limiter, a news feed. The expected depth is assembling standard components with justified choices — a load balancer, application servers, a database choice, a cache — plus rough capacity math. Multi-region failover and deep distributed-systems trade-offs are the senior bar; reaching for them with shaky fundamentals hurts more than it helps. The mid-level-bar module covers this in detail.
After the loop: debrief and hiring committee
You leave the onsite; the decision process is just starting.
Each interviewer submits written feedback with a hire/no-hire recommendation and a level signal. What happens next varies:
- Google: a hiring committee — a standing group of engineers who did not interview you — meets weekly, reviews the whole packet, and decides both hire and level. Coding performance is weighted most heavily in distinguishing L4 from L3.
- Amazon: the debrief includes the Bar Raiser, who has veto power over the hire regardless of what the team wants.
- Meta and most others: the interviewers plus the hiring manager debrief and decide together.
A common outcome for borderline mid-level packets is down-leveling: you interviewed at L4, the offer comes at L3. This is not a rejection and not rare. If it happens, you can sometimes push back with a strong competing offer at level, or ask about re-interviewing for the level after some time.
During this 1–2 week window there is little to actively do. Send a thank-you note through the recruiter, respond fast to any requests for information, and keep your other processes moving — this is exactly the dead time parallel processes are designed to fill.
Team match: the stage nobody warns you about
At Google (and sometimes at Meta, before or after the offer depending on the org), passing the committee does not mean you have a job. It puts you in a pool. Your recruiter proposes teams with open headcount, and you do 30-minute chats with hiring managers to assess mutual fit.
Two things to know:
Your packet can expire. At Google, an approved packet lasts about 8 weeks. If no team picks you in that window — a real risk in slow quarters — the process ends without an offer, after you passed everything.
Managers pick candidates who visibly want their team. Treat each chat as a two-way interview: ask about the team's charter, tech stack, on-call load, growth path, and how the manager supports promotion to senior. Then, for teams you like, express clear enthusiasm — explicitly. "This is exactly the kind of work I want to do, and I'd pick this team" is not overeager; it is how matches happen.
Team match typically takes 2–4 weeks but can stretch to months, which is one more reason to keep parallel processes warm until you have signed something.
The offer
The recruiter delivers a verbal offer — base salary, equity, sign-on bonus, bonus target — then a written one once you signal you are close to accepting. Mid-level total compensation at tier-1 US companies runs roughly $185K–$280K in 2025–2026; the broader non-Big-Tech market runs about $109K–$176K base, with AI/ML-adjacent roles 10–30% higher. Never accept on the call — thank them, ask for the written breakdown, and benchmark it. Almost everything except the base-salary band ceiling is negotiable, and equity and sign-on move the most.