Cognition is an applied AI lab. It makes Devin, an AI agent that carries out software engineering tasks on its own, and Windsurf, an AI code editor that the company renamed Devin Desktop in June 2026. Software engineers there work on agent reasoning, developer tools and agent workflows, so what they ship changes how other engineers write software. The engineering team is small and works closely together; its members include former competitive programmers, people who have founded companies and engineers from other technology and AI firms. Because there are few engineers, each one gets wide autonomy and owns a large share of the product. The role asks you to ship quickly, to make progress on problems that arrive without a clear specification, and to build systems that can work through long, messy tasks from real projects. The pace is demanding, and the job leans every day on your own judgment about what to build and how to build it.
Initial Technical Screen
reportedCandidates describe the first round as a technical screen or an introductory call; for some it included a timed coding exercise.
What to demonstrate
- Working code written under a time limit
- Reading the provided documentation quickly and turning it into working code in a language or environment you may not know
How to prepare
- Answer aloud and timed: Implement a parser for a custom configuration format using a provided set of syntax rules.
- Answer aloud and timed: Write a program to resolve dependencies in a package manager, handling circular dependencies gracefully.
Intensive Technical Interviews
reportedAfter the screen, candidates report several longer technical interviews that combine hands-on coding, algorithm problems and system design.
What to demonstrate
- Hands-on coding and algorithm problems over several long sessions, plus system design
- Reasoning from first principles about data structures, complexity and concurrency, and stating the trade-offs of a design
How to prepare
- Answer aloud and timed: Debug a multi-threaded application where race conditions are causing intermittent failures.
- Answer aloud and timed: Implement an algorithm to optimize file-system traversal, balancing memory usage and execution speed.
Team Conversations
reportedThe last stage candidates describe is a set of talks with different people on the team, focused on how you communicate, how open you are about what you do not know, and how well you fit the way the company works.
What to demonstrate
- Communication, humility about the limits of what you know, and fit with how the team works
- Ownership of ambiguous problems and openness to feedback, across conversations with engineering, product and leadership
How to prepare
- Answer aloud and timed: Build a utility that translates API payloads between two different schema versions based on a dynamic mapping document.
- Answer aloud and timed: Design the backend architecture for an AI-native IDE that requires real-time synchronization of file states and terminal outputs.
PracHub editorial advice for the preparation topics above.
Freezing when the coding test uses a language or framework you do not know
Treat these tests as open book. Look things up in the documentation you are given, say what you are checking and why, and keep moving, because how fast you find and apply the right page is part of what is judged.
Spending the adaptive coding test trying to recall exact syntax from memory
Work out the logic of the solution first, then let the provided documentation settle the syntax as you write each part.
Defending your position instead of listening in the team conversations
Candidates report that maturity and collaboration weigh heavily in the long team interviews. Hold opinions loosely, take feedback without arguing every point, and show how you would work through a disagreement with a colleague.
Choose a category, try a prompt, then open its approach, worked solution or follow-up when you need it.
Implement a parser for a custom configuration format using a provided set of syntax rules.
Implement a parser for a custom configuration format using a provided set of syntax rules.
Approach
- Restate the input: its shape, its size, and what is guaranteed about it.
- Name the brute-force solution and its complexity before improving on it.
- Choose the data structure from the access pattern, not from familiarity.
- State the target complexity and say which constraint rules the naive version out.
Follow-up
- How does this change if the input no longer fits in memory?
- What is the worst case, and how likely is it on real data?
Write a program to resolve dependencies in a package manager, handling circular dependencies gracefully.
Write a program to resolve dependencies in a package manager, handling circular dependencies gracefully.
Approach
- Restate the input: its shape, its size, and what is guaranteed about it.
- Name the brute-force solution and its complexity before improving on it.
- Choose the data structure from the access pattern, not from familiarity.
- State the target complexity and say which constraint rules the naive version out.
Follow-up
- How does this change if the input no longer fits in memory?
- What is the worst case, and how likely is it on real data?
Implement an algorithm to optimize file-system traversal, balancing memory usage and execution speed.
Implement an algorithm to optimize file-system traversal, balancing memory usage and execution speed.
Approach
- Say what the runtime actually does before reasoning about the code.
- Name what is shared across threads and what owns each piece of state.
- Identify the window where an invariant is briefly untrue.
- Distinguish a value from a reference to it, and say which one you handed out.
Follow-up
- What happens if two callers reach this at the same time?
- Where could this allocate more than you expect?
Replace offset paging on the resource feed with keyset
resource holds resource_id, tenant_id, owner_user_id, title, body_ref, version, status ('draft','active','archived','deleted'), created_at, updated_at, deleted_at, with an index on (tenant_id, status, updated_at DESC, resource_id DESC). The listing endpoint returns active resources for one tenant, newest update first, 50 per page, today with LIMIT 50 OFFSET n. Tenants reach page 400 and rows are created while they read. Write the keyset query, define what the cursor carries and how it is encoded, and say which part of the index each predicate uses. Assume PostgreSQL 16.
Approach
- Name the two failures separately. OFFSET 20000 makes the server produce and discard 20,000 rows, so page cost grows with depth rather than with page size. Independently, any write that changes how many rows sort above the offset moves the window between two fetches, and the direction decides which anomaly you get: an insert lands at the head of updated_at DESC and pushes already-returned rows down past the boundary, so they are returned a second time; a delete above the offset, or a row whose updated_at is bumped above the cursor, pulls rows up and one is never returned at all. Nothing in the response reveals either.
- Write the seek: WHERE tenant_id = $1 AND status = 'active' AND (updated_at, resource_id) < ($2, $3) ORDER BY updated_at DESC, resource_id DESC LIMIT 50. The row-value comparison is one index range rather than a disjunction, and both columns are NOT NULL, which is what makes that comparison well defined.
- Map each predicate onto the index: tenant_id and status are equality on the leading columns, (updated_at, resource_id) is the range, and the ORDER BY matches the index order so no Sort node appears and the scan stops after 50 rows. The DESC in the definition only matters for mixed directions — a plain ascending btree on the same columns is read backwards for this query.
- Put both sort columns in the cursor and nothing the client can tamper with into another tenant: base64 of (updated_at, resource_id), validated server-side, with tenant_id taken from the principal.
Follow-up
- The client asks for 'jump to page 400'. What do you offer instead, and what does the honest version cost?
- Sort order becomes user-selectable across four columns. How many indexes is that, and which would you refuse to add?
Denormalise tenant onto revisions and backfill it live
resource_revision (revision_id, resource_id, version, actor_user_id, change_kind, patch, request_id, created_at) has 400M rows and no tenant column; tenant_id lives only on resource. Two reads need it: a tenant-scoped audit feed ordered by created_at DESC, and an offboarding purge. Both join back to resource today. Justify adding tenant_id to resource_revision against those two reads, name the anomaly the copy introduces and the constraint that prevents it, then give the ordered migration for a live table taking 1.2k writes/second — the lock each step takes, how the backfill is batched, and where each step stops being reversible. PostgreSQL 16.
Approach
- Justify from the access path rather than from taste. Without the column, the audit feed either scans resource_revision by created_at and discards other tenants' rows, or resolves the tenant's resource_ids first and probes with them — both proportional to the tenant's whole history rather than to one page. With (tenant_id, created_at DESC, revision_id DESC) it is a seek that stops at 50 rows, and the purge becomes a ranged delete instead of a join.
- Name the cost exactly: a second copy of a fact can disagree with the first. Make the disagreement unwritable rather than documented — add UNIQUE (resource_id, tenant_id) on resource so it can serve as a foreign-key target, then FOREIGN KEY (resource_id, tenant_id) REFERENCES resource (resource_id, tenant_id) on the revision table. A revision can then only ever carry its parent's tenant.
- Step one, expand: ALTER TABLE resource_revision ADD COLUMN tenant_id BIGINT NULL, with no default, so it is a catalogue change and no rewrite. It still needs ACCESS EXCLUSIVE for an instant, and that instant queues behind the longest open transaction on the table while every later query queues behind it — set lock_timeout to 2s and retry rather than wait.
- Step two, dual-write: deploy the writer that populates tenant_id on every new revision while reads still use the join. Reversible by redeploying the previous build, because nothing reads the column yet.
Follow-up
- The backfill is half finished and a rollback is required. What state is the table in, and what does the previous build do with a half-populated column?
- How do you verify the backfill actually finished, given rows are still being inserted while it runs?
Build a utility that translates API payloads between two different schema versions based on a dynamic mapping
Build a utility that translates API payloads between two different schema versions based on a dynamic mapping document.
Approach
- Say who the caller is and what they do when the call fails halfway.
- Define the identity of a request so a retry cannot double-apply it.
- Separate accepted, pending, failed and confirmed; they are different facts.
- Design the error taxonomy before the success shape; callers branch on it.
Follow-up
- What happens if the caller retries after a timeout?
- How does a client discover it is on an old version of this contract?
Design the backend architecture for an AI-native IDE that requires real-time synchronization of file states an
Design the backend architecture for an AI-native IDE that requires real-time synchronization of file states and terminal outputs.
Approach
- Fix the scope first: who calls this, how often, and what they do when it fails.
- Name the read and write paths separately; they rarely have the same bottleneck.
- Choose a partition key and say what query it makes expensive.
- State the consistency you need, and where you are willing to be stale.
Follow-up
- What breaks first when traffic grows ten times?
- How does this behave when that dependency is down for an hour?
How would you architect a secure, isolated sandbox environment for executing untrusted code generated by an AI
How would you architect a secure, isolated sandbox environment for executing untrusted code generated by an AI agent?
Approach
- Fix the scope first: who calls this, how often, and what they do when it fails.
- Name the read and write paths separately; they rarely have the same bottleneck.
- Choose a partition key and say what query it makes expensive.
- State the consistency you need, and where you are willing to be stale.
Follow-up
- What breaks first when traffic grows ten times?
- How does this behave when that dependency is down for an hour?
Design a rate-limiting and queuing system for an API that experiences highly variable, bursty traffic from ent
Design a rate-limiting and queuing system for an API that experiences highly variable, bursty traffic from enterprise customers.
Approach
- Fix the scope first: who calls this, how often, and what they do when it fails.
- Name the read and write paths separately; they rarely have the same bottleneck.
- Choose a partition key and say what query it makes expensive.
- State the consistency you need, and where you are willing to be stale.
Follow-up
- What breaks first when traffic grows ten times?
- How does this behave when that dependency is down for an hour?
How would you design an integration layer that connects a proprietary AI agent with a client's legacy database
How would you design an integration layer that connects a proprietary AI agent with a client's legacy database and internal APIs?
Approach
- Fix the scope first: who calls this, how often, and what they do when it fails.
- Name the read and write paths separately; they rarely have the same bottleneck.
- Choose a partition key and say what query it makes expensive.
- State the consistency you need, and where you are willing to be stale.
Follow-up
- What breaks first when traffic grows ten times?
- How does this behave when that dependency is down for an hour?
How do you balance the trade-off between shipping quickly to gather user feedback and spending time on archite
How do you balance the trade-off between shipping quickly to gather user feedback and spending time on architectural polish?
Approach
- Fix the scope first: who calls this, how often, and what they do when it fails.
- Name the read and write paths separately; they rarely have the same bottleneck.
- Choose a partition key and say what query it makes expensive.
- State the consistency you need, and where you are willing to be stale.
Follow-up
- What breaks first when traffic grows ten times?
- How does this behave when that dependency is down for an hour?
Debug a multi-threaded application where race conditions are causing intermittent failures.
Debug a multi-threaded application where race conditions are causing intermittent failures.
Approach
- Establish what changed and when, before forming any theory.
- Pick a bisection that eliminates candidates whichever way it turns out.
- Check the instrumentation before believing the symptom.
- Separate the trigger from the cause; the deploy is rarely the bug.
Follow-up
- What would you look at first, and what would it rule out?
- How would you tell a cause from a coincidence here?
Describe how you would monitor, log, and debug distributed executions of long-running autonomous workflows.
Describe how you would monitor, log, and debug distributed executions of long-running autonomous workflows.
Approach
- Establish what changed and when, before forming any theory.
- Pick a bisection that eliminates candidates whichever way it turns out.
- Check the instrumentation before believing the symptom.
- Separate the trigger from the cause; the deploy is rarely the bug.
Follow-up
- What would you look at first, and what would it rule out?
- How would you tell a cause from a coincidence here?
Built from the rounds and topics Cognition candidates report.
Prepare, practise & reflect
One practical outcome each day. Spend longer where you need it.
0 / 7 done01Map the Cognition loop
- Write out the reported sequence: Initial Technical Screen, Intensive Technical Interviews, Team Conversations.
- For each round, write one sentence on what it is judging, from the description above, and mark the one you are least ready for.
Deliverable: A one-page map of the 3 reported rounds, with the weakest marked.
02Code from documentation in an unfamiliar language
- Choose a small open-source library in a language you have never shipped code in, keep only its reference docs open, and time yourself making one change until its test suite is green.
- Write down each thing you had to look up and how long it took, so you know where the time went.
Deliverable: One timed change to an unfamiliar library with its unit tests passing, and the list of lookups it needed.
03Algorithms and concurrency
- Write a function that reads a stream of execution log lines once, filters and aggregates them while holding as little in memory as possible, and state its time and space complexity aloud.
- Write a small program whose threads share state, make it fail with a race condition, then fix it and explain why the fix holds.
Deliverable: One timed streaming solution with its complexity stated, and one race condition you reproduced and fixed.
04System design for AI agents
- Sketch an integration layer that lets an AI agent read from and write to a client's legacy database and internal APIs: what it may touch, how a retried call stays safe, and what it records.
- List the logs and metrics you would need to debug a long-running autonomous workflow that fails halfway through.
Deliverable: One integration-layer sketch with its access limits, its retry behaviour and the signals you would monitor.
05Answer out loud: Coding & Adaptability
- Answer aloud, timed: Implement a parser for a custom configuration format using a provided set of syntax rules.
- Answer aloud, timed: Write a program to resolve dependencies in a package manager, handling circular dependencies gracefully.
Deliverable: Spoken answers to 2 reported Coding & Adaptability question(s), under time.
06Answer out loud: System Design & Architecture
- Answer aloud, timed: Design the backend architecture for an AI-native IDE that requires real-time synchronization of file states and terminal outputs.
- Answer aloud, timed: How would you architect a secure, isolated sandbox environment for executing untrusted code generated by an AI agent?
Deliverable: Spoken answers to 2 reported System Design & Architecture question(s), under time.
07Answer out loud: Behavioral & Alignment
- Answer aloud, timed: Tell me about a time you had to deliver a highly critical feature with minimal requirements and a tight deadline. How did you decide what to build?
- Answer aloud, timed: Describe a situation where you had a strong disagreement with a teammate over an architectural decision. How did you resolve it?
Deliverable: Spoken answers to 2 reported Behavioral & Alignment question(s), under time.
Expand any day for tasks and deliverables. Your progress is saved on this device.
Behavioural rounds judge the decision you made and what it cost.
Tell me about a time you had to deliver a highly critical feature with minimal requirements and a tight deadli
Tell me about a time you had to deliver a highly critical feature with minimal requirements and a tight deadline. How did you decide what to build?
Approach
- Pick a story where you made the decision, not one where you watched it.
- State the situation in two sentences and spend the rest on the reasoning.
- Give the blast radius: what could have broken, and what you measured.
- Name the disagreement and how you resolved it with evidence.
Follow-up
- What would you do differently if you ran that again?
- How did you know your change caused the improvement?
Describe a situation where you had a strong disagreement with a teammate over an architectural decision. How d
Describe a situation where you had a strong disagreement with a teammate over an architectural decision. How did you resolve it?
Approach
- Pick a story where you made the decision, not one where you watched it.
- State the situation in two sentences and spend the rest on the reasoning.
- Give the blast radius: what could have broken, and what you measured.
- Name the disagreement and how you resolved it with evidence.
Follow-up
- What would you do differently if you ran that again?
- How did you know your change caused the improvement?
Talk about a project you owned from end to end. What was the impact, and what would you do differently today?
Talk about a project you owned from end to end. What was the impact, and what would you do differently today?
Approach
- Pick a story where you made the decision, not one where you watched it.
- State the situation in two sentences and spend the rest on the reasoning.
- Give the blast radius: what could have broken, and what you measured.
- Name the disagreement and how you resolved it with evidence.
Follow-up
- What would you do differently if you ran that again?
- How did you know your change caused the improvement?
Tell me about a complex technical problem you solved that was outside of your comfort zone. How did you ramp u
Tell me about a complex technical problem you solved that was outside of your comfort zone. How did you ramp up?
Approach
- Pick a story where you made the decision, not one where you watched it.
- State the situation in two sentences and spend the rest on the reasoning.
- Give the blast radius: what could have broken, and what you measured.
- Name the disagreement and how you resolved it with evidence.
Follow-up
- What would you do differently if you ran that again?
- How did you know your change caused the improvement?
- 01
Tell me about a time you had to deliver a highly critical feature with minimal requirements and a tight deadline. How did you decide what to build?
- 02
Describe a situation where you had a strong disagreement with a teammate over an architectural decision. How did you resolve it?
- 03
Talk about a project you owned from end to end. What was the impact, and what would you do differently today?
- 04
Tell me about a complex technical problem you solved that was outside of your comfort zone. How did you ramp up?
How hard is Cognition's technical interview?
Candidate reports put the difficulty between average and hard. Solid algorithm and coding skills are expected, and so is the ability to pick up an unfamiliar language or tool while the test is running.
Cognition Software Engineer candidate reports ↗Can Cognition engineers work remotely?
Mostly not. Most engineering roles are reported to be in person, five days a week, at the San Francisco office, because the company favours working side by side with short feedback loops between teammates.
Cognition Software Engineer candidate reports ↗What does Cognition look for most in a candidate?
Initiative paired with humility. Candidates describe the ideal hire as a builder who takes on hard problems without waiting to be told, and who still works easily, without ego, inside a small and close team.
Cognition Software Engineer candidate reports ↗How do I practise for coding tests in a language I do not know?
Solve practice problems in a language or library you have not used, with only its official documentation open. Each time, try to cut how long it takes you to find the right reference and turn the requirements into code that runs.
Cognition Software Engineer candidate reports ↗Which topics come up most in Cognition interviews?
Cognition interviews most often cover three technical areas: coding in a language or environment you may not know, using the documentation you are given; algorithms, data structures and concurrency; and the design of systems around AI agents, such as a sandbox for AI-generated code, an integration layer to a client's legacy systems and monitoring for long-running autonomous workflows. How much weight each area gets varies with the role.
Cognition Software Engineer candidate reports ↗Sources & methodology 3 sources ↗
No official company page is cited. Rounds and questions come from candidate reports and PracHub editorial material; each source shows the date it was read.
- 01Cognition Software Engineer candidate reports ↗
Company-reported rounds, questions and FAQ.
Candidate reports · Accessed 2026-09-22 - 02PracHub Software Engineer practice ↗
PracHub practice material, not company-reported.
PracHub page · Accessed 2026-09-22 - 03PracHub preparation framework ↗
PracHub preparation guidance.
PracHub page · Accessed 2026-09-22
Explore related engineering environments and compare the preparation focus.
Viber Software Engineer Interview Guide
What Viber asks Software Engineer candidates, from reported rounds and questions.
The Coca-Cola Software Engineer Interview Guide
What The Coca-Cola asks Software Engineer candidates, from reported rounds and questions.
Red Ventures Software Engineer Interview Guide
What Red Ventures asks Software Engineer candidates, from reported rounds and questions.
INDmoney Software Engineer Interview Guide
What INDmoney asks Software Engineer candidates, from reported rounds and questions.