Software developer interview questions fall into five predictable rounds, and each round scores something different: pattern recognition in code, architectural judgement, verification habits, and how clearly you explain a decision out loud.
The pipeline of open roles has not dried up — the U.S. Bureau of Labor Statistics projects 10% growth for software developer jobs between 2025 and 2035. What changed is the format: two of the largest employers now run a coding round where you are expected to work alongside an AI assistant.
This guide lists the questions you will actually be asked, round by round, and gives you a reusable framework for answering each type. For the wider process — recruiter calls, scheduling, offer stage — read our complete technical interview guide.
What Questions Are Asked in a Software Developer Interview?
Almost every process, from a 30-person startup to a FAANG loop, is assembled from the same five rounds. Knowing which round you are sitting in tells you what the interviewer is scoring and how long your answer should run.
- 1
Recruiter screen
Motivation, salary range and a walk-through of your CV. No code. The real question is whether your experience matches the level the team is hiring for.
20-30 min
- 2
Technical phone screen
One or two medium problems in a shared editor, plus language fundamentals. Narrating your approach counts as much as finishing the code.
45-60 min
- 3
Coding round
Two problems, or one problem you extend twice. At Meta and Google this is where an AI assistant may now be part of the exercise.
60 min
- 4
System design
Mandatory from mid-level. An open prompt such as Design a chat application, scored on scoping, trade-offs and failure modes.
45-60 min
- 5
Behavioural round
Conflict, ownership, failure and influence, answered as structured stories with numbers rather than as opinions.
45 min
Prepare by round, not by topic. Candidates who grind 200 algorithm problems and skip story preparation still fail at the behavioural stage, because that round tests a skill the coding round never touches.
Coding Questions: Patterns, Examples and the REACTO Framework
Most coding questions are variations on six recurring patterns. Recognising the pattern in the first minute is worth more than memorising individual solutions, because the interviewer picks variants you have never seen.
| Pattern | Example question | What it signals |
|---|---|---|
| Two pointers | Find the pair that sums to a target in a sorted array | You notice the input is sorted before reaching for extra memory |
| Sliding window | Longest substring without repeating characters | You can carry state across a moving range instead of re-scanning |
| Hash map counting | Group anagrams, first non-repeating character | You trade memory for time deliberately, not by reflex |
| BFS and DFS | Number of islands, word ladder | You pick the traversal order from the question, not from habit |
| Heap and top-K | Top K frequent elements, median from a data stream | You know when sorting the whole input is wasteful |
| Dynamic programming | Coin change, longest common subsequence | You can define a state and a transition before writing code |
The REACTO framework for any coding question
Use the same six steps on every problem. It keeps you talking when you are stuck, which is exactly when interviewers take notes.
- Repeat: restate the problem in your own words and confirm inputs, outputs and constraints.
- Examples: write one normal case and two edge cases before any code.
- Approach: describe the brute force, give its complexity, then name the optimisation you will implement.
- Code: write it while narrating each decision, not in silence.
- Test: run your own examples by hand, line by line, and fix what breaks.
- Optimise: state final time and space complexity plus one thing you would change with more time.
Real coding questions by experience level
Entry level (0-2 years): Two Sum; Valid Parentheses; merge two sorted lists; reverse a linked list; find the missing number in a range; check whether a string is a palindrome.
Mid level (2-5 years): longest substring without repeating characters; binary tree level-order traversal; number of islands; implement an LRU cache; merge overlapping intervals; validate a binary search tree.
Senior (5+ years): serialise and deserialise a binary tree; course schedule with topological sort; median from a data stream; implement a rate limiter in code; word ladder; refactor and optimise a function the interviewer hands you.
The AI-Assisted Round Reshaped Software Developer Interview Questions in 2026
Meta began rolling out an AI-enabled coding round in October 2025, replacing one of the two traditional onsite coding interviews. Google is piloting a comparable format with Gemini for junior and mid-level roles on selected US teams, as documented in this university career-office breakdown of the AI-assisted coding interview.
Speed is not what these rounds score. Meta grades on four criteria — problem solving, code quality, verification and communication — while Google evaluates AI fluency: how precisely you scope a prompt, how you validate the output, and how you debug what comes back.
That emphasis on verification mirrors what developers already report: adoption of these tools has run far ahead of trust in them, and debugging generated code is the single most cited frustration in the 2025 Stack Overflow Developer Survey.
Expect prompts built around reading and correcting code rather than writing it from a blank file:
- Read this unfamiliar function and explain what it does before you change anything.
- Ask the assistant for an implementation, then tell us which parts you would refuse to ship.
- The generated solution fails on this input — diagnose it out loud.
- Extend this small codebase with a new endpoint and justify where you placed it.
Only use an assistant when the company states that the round allows it. Formats are not converging: the same employer may run one AI-enabled round and one AI-free round in the same loop, and using a tool in the wrong round ends the process.
System Design Questions for Mid-Level and Senior Developers
System design is where levelling decisions are made. The prompt is deliberately vague, and the first thing scored is whether you narrow it before designing anything.
| Prompt | What the interviewer probes |
|---|---|
| Design a URL shortener | Key generation, collisions, read-heavy caching |
| Design a chat application | Delivery guarantees, presence, ordering of messages |
| Design a news feed | Ranking, fan-out on write versus on read, cold start |
| Design a rate limiter | Token bucket versus sliding window, shared state |
| Design a file storage service | Chunking, deduplication, consistency model |
| Design a notification system | Queues, retries, idempotency, delivery windows |
Structure the 45 minutes in five blocks: clarify the scope (5 minutes), estimate load and storage (5 minutes), draw the high-level components (10 minutes), go deep on the one component the interviewer points at (15 minutes), and close on trade-offs and failure modes (10 minutes).
The difference by level is the depth, not the prompt. A junior candidate is expected to place the components correctly; a senior candidate is expected to name what breaks first under ten times the load, and what they would sacrifice to keep the system up.
Why Do Strong Engineers Fail the Behavioural Round?
They answer with opinions instead of episodes. Asked how they handle disagreement, they describe a philosophy; the interviewer needed a dated situation, a decision that was theirs, and a measurable outcome.
Fix it with a structure. Spend about half the answer on the actions you took, and always end on a number or a concrete consequence — the mechanics are covered in our guide to structuring an interview answer, with eight full worked responses in these STAR method examples.
The behavioural questions developers actually get
- Conflict: Tell me about a technical decision you disagreed with. What did you do? How did it end?
- Ownership: Describe a project you owned end to end. What would you build differently now?
- Failure: Tell me about a bug that reached production because of something you missed.
- Influence: Describe a time you convinced a team to change an approach without any authority to impose it.
- Ambiguity: Tell me about a task where the requirements kept moving. How did you decide what to build?
- Learning: Describe the last technology you had to learn under deadline pressure.
Prepare six stories, not sixty. Six episodes with real numbers can be re-framed to answer almost any behavioural prompt, and rehearsed stories stay under two minutes.
Language-Specific Questions: JavaScript, Python, Java and SQL
Most loops include one round of language fundamentals, usually the first fifteen minutes of the technical screen. These questions are checkable facts, so they are the cheapest points in the whole process.
JavaScript: explain closures with a practical example; the difference between var, let and const; how the event loop orders microtasks and macrotasks; implement debounce from scratch.
Python: lists versus tuples and when the distinction matters; what the Global Interpreter Lock does to threading; how decorators work; write a context manager both as a class and as a generator.
Java: ArrayList versus LinkedList in real workloads; how generational garbage collection works; abstract classes versus interfaces since default methods; implement a thread-safe singleton.
SQL: the difference between INNER JOIN and LEFT JOIN on missing rows; when an index makes a query slower; what a transaction isolation level changes; find duplicate rows in one query.
A Four-Week Preparation Plan
Four weeks at ninety minutes a day is enough for most mid-level loops, provided the time is split across rounds instead of spent entirely on algorithms.
- Week 1 — patterns: three problems a day from the six patterns above, always narrated out loud with REACTO.
- Week 2 — design and language: two system design prompts fully timed, plus the fundamentals list for your main language.
- Week 3 — stories: write your six behavioural episodes, each with a number in the result, and cut every one to two minutes.
- Week 4 — full rehearsals: simulate complete rounds under time, including one session where you deliberately work with an AI assistant and verify everything it produces.
Rehearsal is the part candidates skip, and it is the part that shows. If you want timed practice with feedback on how your answers actually land, our AI interview simulator runs the same rounds described here.
Frequently Asked Questions
How many coding questions should I practise before an interview?
Around 50 to 60 problems is enough if they are chosen to cover the six patterns rather than picked at random. Depth beats volume: solving the same problem twice, the second time without notes, teaches more than a new problem you half-finish.
Are you allowed to use AI in a coding interview?
Only in rounds where the employer explicitly permits it, such as Meta's AI-enabled onsite round or Google's Gemini pilot. In every other round, assume it is prohibited and treat any undisclosed use as grounds for rejection.
What should I do when I do not know the answer?
Say where your knowledge ends, then reason forward from what you do know. Interviewers score the recovery: a candidate who narrows the problem and proposes a way to verify an assumption outperforms one who guesses confidently.
Do junior developers get system design questions?
Increasingly yes, but in a reduced form — usually a single service such as a URL shortener, judged on whether you place components sensibly. Deep scaling and consistency trade-offs remain a mid-level and senior expectation.
Which programming language should I use in the coding round?
The one whose standard library you know without looking things up, unless the job description names a language explicitly. Fluency in the idioms of one language reads better than shaky code in the language you think they want.