Cognition interview, decoded.
Updated · Sources at the end
Cognition's CEO, Scott Wu, has described the company's engineering interview in one line: people build their own Devin in eight hours, and the team sees how far they get. Devin is Cognition's autonomous coding agent, so the exercise is the product itself, rebuilt from nothing under a clock. Most of the company's open engineering roles are customer-facing forward deployed engineers (FDEs), and their loop is written up in more detail: a take-home, a presentation of a past project, conversations with leadership, a pitch to a panel playing executives, and a live customer call. Candidates in September 2026 describe the take-home as an automation built on the Devin API, with a five-minute video pitch, due a few days later.
Cognition makes Devin and, since buying it in July 2025, the Windsurf editor. On 8 September 2026 it announced a Series E of more than $2B at a $48B valuation, and said its run-rate revenue had grown from $492M in May to almost $900M.
Free · 2 minutes · no account
The questions for your exact Cognition role and level.
Who Cognition hires
Most open engineering roles are in customer engineering: forward deployed engineers, applied AI engineers, partner deployed engineers and AI support engineers, in San Francisco, Washington DC, London, Southern Europe, Tokyo, Singapore, Sydney, Seoul, São Paulo, Mexico City, India and Doha. On 11 November 2026 the board listed 25 customer engineering roles of 116, including a partner deployed engineer for Europe in London posted on 6 November. The research and development team hires software, infrastructure, data and security engineers, a product engineer, and research engineers for post-training, mid-training, ML infrastructure and, since October 2026, safety and alignment, all in San Francisco. The ads say the founding team includes competitive programmers and people from Scale AI, Palantir, Cursor, Waymo, Tesla and Google DeepMind.
Wondering whether your resume reads like someone Cognition hires? Get a free resume review.
Two loops are on the record. The build exercise belongs to the core engineering team; the deployed engineer loop is about getting Devin adopted inside a customer.
Track · R&D
Software and research engineering
The engineering interview Scott Wu described in August 2025: build your own Devin in eight hours. The software engineer ad describes the work the exercise stands for: tool use, context management, multi-step planning, subagents and sandboxed code execution, shipped into Devin and Windsurf. A software engineer reported in October 2026 was asked about an end-to-end product they worked on.
Track · FDE
Deployed engineering
Deployed, applied AI and partner deployed engineers embed with a customer's engineers and get Devin into real production work. The loop: recruiter screen, a take-home, a 45-minute presentation of a past project, one-on-ones with leadership, a pitch to a mock executive panel, and a case-study call with a simulated customer. The take-home reported in September 2026 is an event-driven automation on the Devin API with a five-minute video pitch; earlier reports describe a lighter exercise inside Devin with little code.
What is the Cognition interview process?
The deployed engineer loop below comes from the guide by Aced (formerly Exponent), last updated in October 2026, and from candidate reports posted there between August and October 2026. The build exercise comes from the CEO himself, in August 2025, when core engineering was about 19 people; the team has grown since, so ask your recruiter what the build looks like now.
Your background, what you know about Cognition, and for deployed roles why forward deployed engineering. Deployed candidates are judged here on explaining technical work to someone who is not technical; one in September 2026 was also asked what to do when the largest customer wants a feature that is not on the roadmap. Some report a second call that digs into their current role and their customers.
Build an agent that writes and runs code, then show how far it got. Candidates are measured on how far they get, so a working loop early beats a clever design that never runs.
EngineeringReported in September 2026: build an event-driven automation on the Devin API that solves an engineering workflow problem, then record a five-minute video pitching it. Candidates call the timeline tight on purpose. Earlier accounts, and Aced's guide, describe a lighter exercise inside Devin with very little code; either way it tests whether you understand the product the way a customer will need to.
Deployed engineeringA past project, told for people outside your field: why it mattered in the first five minutes, then the decisions, what you owned, and how you handle follow-up questions.
Deployed engineeringWhat motivates you, a project you owned end to end, and how steady you stay when someone senior pushes back.
Sell Devin in business terms: faster delivery, lower cost, more ambitious work. Features only count when they connect to an outcome.
Deployed engineeringStructure an unfamiliar customer problem quickly, then adapt when the customer's questions do not follow your plan.
Deployed engineeringWhat Cognition screens for
The ads for every engineering role list the same things, and the deployed engineer loop is built to test them in front of a customer.
- Customer empathy: how engineers actually work and what helps them
- Ownership of the outcome a customer sees
- Technical curiosity about unfamiliar systems
- A bias toward execution, from problem to working solution quickly
- Fluency with agents: how you use them, and how you would build one
Cognition interview questions
Candidate-reported themesThese are the questions reported on Aced (formerly Exponent) for Cognition, most of them for deployed engineers. The engineering ones are recent and few, which matches a loop built around one long exercise.
Behavioural & motivation
- Why forward deployed engineering?
- Walk me through a technical feature you led.
- How much are you owning on a deployment?
- Present a past technical project to a non-technical audience.
- Your largest customer is advocating for a new feature that is not on your roadmap. What do you do?
- How would you rank yourself in terms of percentile?
- Tell me about yourself.
Technical
- You have eight hours to build your own Devin, a coding agent from nothing. What do you build first, and how do you extend it?Build an agent: A coding agent from nothing: a model in a loop with tools, a sandbox to run code, and a way to plan multi-step workListening for: Working loop early · Key parts · Sensible extension
- Build an event-driven automation on the Devin API that solves an engineering workflow problem.The Devin API: The deployed engineer take-home reported in September 2026, followed by a five-minute video pitchListening for: Real workflow · Event driven design · Reliability
- How do you use agents in your own work, and where do they fail?Agentic development: How you use agents in your own work, and where they failListening for: Concrete usage · Where they fail · How you manage it
- Walk me through the technical architecture of the product you work on now.Your product's architecture: Reported for software engineers: walk through the technical architecture of the product you work on nowListening for: Clear overview · Key decisions · Your part
- Practice: A customer wants Devin working in their repositories, but their security team restricts network access and secrets. How do you get it into real production work?Deployment and integration: For deployed engineers: getting an agent working inside a customer's repositories, tools and security rulesListening for: Discovery first · Security rules · Adoption and outcome
These are the reported themes, and your loop is role-specific. Paste the actual posting and Calibrd predicts the questions for that exact role and level. How the questions are predicted →
Predict my Cognition questions →Pay
Cognition posts ranges on only some ads, read on 23 September 2026 and unchanged on 11 November 2026: software, infrastructure and security engineers in research and development at $260K to $300K base plus significant early-stage equity, applied AI engineers at $180K to $225K base plus equity, and federal deployed engineers at $180K to $280K base plus equity and a bonus. The other deployed engineer ads carry no range. Levels.fyi's 'Cognition' pages mix in other companies with the same name, so they are not used here. Equity is the large part of the package at a company valued at $48B in September 2026. Calibrd benchmarks the pay on any posting you scan against its level and location.
Applying to Cognition? Paste the posting and see whether they’d shortlist you →
How to prepare for a Cognition interview
- Build a coding agent before the interview: a model calling tools in a loop, running code in a sandbox, reading the output and trying again. Get it working end to end in the first couple of hours, then extend it.
- Use Devin. The deployed take-home happens inside it, and every loop assumes you know what it does and where it struggles.
- Deployed engineers: prepare the project presentation so the impact lands in the first five minutes, and rehearse a pitch that ties Devin to delivery speed and cost.
- Practise a customer call where the customer goes off script. Structure fast, then let it change.
- Have a real answer on agentic development: what you use, what you built with it, where it broke.
This guide covers Cognition's engineering and deployed engineering loops. Sales, marketing and operations roles run a different process. For management and leadership roles the loop is similar but the bar shifts to people, delivery and strategy, so pair it with the leadership interview prep hub. The bar for your exact role comes from the role-by-role guides, and the prep that actually transfers is spoken, so run a mock interview before the real one.
Also interviewing at Synthesia, Databricks or Firecrawl? Their guides follow the same format, from the first screen to the offer.
Knowing the questions isn’t the same as answering them out loud. Run a Cognition mock: spoken answers, coached on the spot. Your first mock is free.
Run a Cognition mock →FAQ & sources
What is Cognition's interview process?
The build exercise comes from the CEO himself, in August 2025, when core engineering was about 19 people; the team has grown since, so ask your recruiter what the build looks like now. The stages, in order: Recruiter screen; Build your own Devin; Take-home; Project presentation; Leadership one-on-ones; Executive pitch; Customer case call.
What does Cognition look for in candidates?
The ads for every engineering role list the same things, and the deployed engineer loop is built to test them in front of a customer. Customer empathy; Ownership of the outcome a customer sees; Technical curiosity about unfamiliar systems; A bias toward execution, from problem to working solution quickly.
What questions does Cognition ask in interviews?
These are the questions reported on Aced (formerly Exponent) for Cognition, most of them for deployed engineers. Technical themes: Build an agent; The Devin API; Agentic development; Your product's architecture; Deployment and integration.
How much does Cognition pay?
Software Engineer, R&D (posted): $260K–$300K base plus significant early-stage equity, San Francisco; Applied AI Engineer (posted): $180K–$225K base plus equity, SF, Singapore, Sydney; Deployed Engineer, Federal (posted): $180K–$280K base plus equity and bonus, Washington DC. Cognition posts ranges on only some ads, read on 23 September 2026 and unchanged on 11 November 2026: software, infrastructure and security engineers in research and development at $260K to $300K base plus significant early-stage equity, applied AI engineers at $180K to $225K base plus equity, and federal deployed engineers at $180K to $280K base plus equity and a bonus.
How do I get a job at Cognition?
Most open engineering roles are in customer engineering: forward deployed engineers, applied AI engineers, partner deployed engineers and AI support engineers, in San Francisco, Washington DC, London, Southern Europe, Tokyo, Singapore, Sydney, Seoul, São Paulo, Mexico City, India and Doha. On 11 November 2026 the board listed 25 customer engineering roles of 116, including a partner deployed engineer for Europe in London posted on 6 November. Hiring runs in 7 stages, starting with Recruiter screen.
How do I prepare for a Cognition interview?
Build a coding agent before the interview: a model calling tools in a loop, running code in a sandbox, reading the output and trying again. Get it working end to end in the first couple of hours, then extend it. Use Devin. The deployed take-home happens inside it, and every loop assumes you know what it does and where it struggles.
- 01Cheeky Pint, Cognition CEO Scott Wu on acquiring WindsurfScott Wu: 'our whole interview process... is basically just having people build their own Devin in eight hours, and seeing how far they get with it', and core engineering at about 19 people growing to 30 to 35 after Windsurf; 27 August 2025
- 02Aced (formerly Exponent), Cognition forward deployed engineer interviewthe six stages (recruiter screen, take-home inside Devin with minimal coding, 45-minute project presentation, leadership one-on-ones, executive pitch, customer case call) and that the loop drops the coding and system design rounds; updated October 2026, read 11 November 2026
- 03Aced (formerly Exponent), Cognition interview questions19 reported questions on 11 November 2026, including the Devin API automation and agentic development questions for forward deployed engineers (September 2026) and the end-to-end product question for software engineers (October 2026)
- 04Aced, Cognition forward deployed engineer interview (Singapore, mid level)recruiter screen with the roadmap feature and agentic development questions, then the Devin API take-home with a five-minute video pitch, sent a day or two after the call with a few days to finish; posted September 2026
- 05Aced, Cognition forward deployed engineer interview (Brazil)recruiter screen ('Tell me what you know about Cognition'), then a phone interview on the current role, team and customers; posted September 2026
- 06Aced, Cognition forward deployed engineer interview (offer)recruiter screen, a light product take-home ('not a leetcode round'), a project presentation, then executive pitch and customer scenarios in the final round; posted August 2026
- 07Cognition careers, Software Engineeragent infrastructure (tool use, context management, planning, subagents, sandboxed execution), work on Devin and Windsurf; read 23 September 2026
- 08Cognition careers, Deployed Engineerthe customer-facing role and what the company values: customer empathy, ownership, technical curiosity, execution; read 23 September 2026
- 09Cognition job board116 open roles on 11 November 2026 (101 on 23 September), 25 in customer engineering and 11 in research and development, the role titles now forward deployed engineer, the partner deployed engineer for Europe posted 6 November and the safety and alignment research engineer posted 9 October, with the posted ranges for software, infrastructure, security, applied AI and federal forward deployed engineers
- 10Cognition, Do it all with Devin: announcing our Series Emore than $2B at a $48B post-money valuation, led by Andreessen Horowitz and Accel; run-rate revenue from $492M to almost $900M since May; offices in Washington DC, Tokyo, Singapore, London, São Paulo and Madrid; 8 September 2026
- 11TechCrunch, Cognition raises $1B at a $25B pre-money valuationthe May 2026 round led by Lux Capital, General Catalyst and 8VC, $26B post-money, $492M annualised revenue
- 12Bloomberg, Cognition raises $2 billion at a $48 billion valuethe Series E, 8 September 2026
- 13TechCrunch, Cognition offers Windsurf staff the exit doorthe Windsurf acquisition in July 2025 and the buyouts that followed
Interview processes change. This reflects widely-reported and sourced conditions as of 2026 — confirm specifics with your recruiter, and treat it as a map rather than a guarantee.
Prep for a real Cognition role
Practise your Cognition interview, out loud.
Paste a real Cognition posting and Calibrd predicts the questions for that role and level, benchmarks the pay, and flags the gaps an interviewer will probe in your resume — then listens to your spoken answers and coaches them. Your first mock is free.
Free to start · No card · Encrypted at rest, never used to train AI, remove anytime