Decagon interview, decoded.

Updated · Sources at the end

Decagon builds AI agents that answer customer service conversations for large companies, and its engineering interviews are built from the same material. Candidates who posted their loops in 2026 describe a hiring call, a 45-minute coding screen with a single question, a system design screen, and an onsite. The coding problems recur from one report to the next: a class that tracks customer satisfaction (CSAT) scores, a shopping cart that keeps versions so you can undo and redo, and a count of every legal game of tic-tac-toe. The design rounds ask for systems that look like Decagon's own product: a place where customers configure rules for their AI agents, a runtime that runs those agents, and the security around a chatbot. One candidate was told the round was system design and found it was about data models, APIs and data flow, with scale left out.

Decagon was founded by Jesse Zhang and Ashwin Sreenivas. In January 2026 it raised $250 million at a $4.5 billion valuation, which it says tripled its valuation in under six months. A summary of a talk Zhang gave in September 2026 says the team grew about tenfold in a year, through more than 2,000 interviews, and that the founders personally met every candidate until the company reached about 300 people. Every posting says it is an in-office company, with most roles in San Francisco, New York and London.

Free · 2 minutes · no account

The questions for your exact Decagon role and level.

01

Who Decagon hires

On 14 November 2026 Decagon had 124 open postings on its Ashby board: 37 in sales, 28 in engineering, 23 on the customer team, 10 deployment strategists, and the rest in finance, the founders' office, people, marketing, product and university hiring. By location, 64 were in San Francisco, 23 in New York and 12 in London, with a few in Australia, Toronto, Austin, Amsterdam and remote. Engineering hires senior and staff software engineers for product, the agent platform, voice, security, cloud and data infrastructure, plus research engineers and a new grad Member of Technical Staff role. Around them sit two roles that exist because of the product: Agent Deployment Engineers, whom Decagon describes as customer engineers who own the technical agent build, and Agent Product Managers, whom the deployment engineers work alongside. Every posting carries the same line: "We're an in-office company, driven by a shared commitment to excellence and velocity."

Wondering whether your resume reads like someone Decagon hires? Get a free resume review.

Engineers and the customer-facing agent roles go through different interviews. The engineering loop is the one candidates have written up in detail; the agent roles are known only from short reviews.

  • Track · Eng

    Software and AI engineering

    A hiring call, a coding screen, a system design screen and an onsite, in the 2026 reports. The coding is practical and object-oriented, with follow-ups that add a requirement to the class you just wrote. The design is product-shaped: agent rules, an agent runtime, chatbot security. An AI Engineer candidate got the same tic-tac-toe problem as the software engineers.

  • Track · Agent

    Agent deployment and agent product

    Agent Deployment Engineers own the technical build of each customer's agent, working alongside Agent Product Managers and the customer. Candidates for the Agent PM role describe a take-home case study followed by a presentation to other Agent PMs, but those reviews sit on Glassdoor and could only be read in search summaries, so treat the shape as unconfirmed and ask your recruiter. Posted pay for an Agent Deployment Engineer is $175K to $230K in San Francisco and New York.

02

What is the Decagon interview process?

Decagon publishes no interview process: its careers page and all 124 postings read on 14 November 2026 are silent on it. What follows comes from five first-hand reports on PracHub (interviews from April to October 2026) and one on Aced, each marked as a candidate's account. They agree on the order and on the problems, and they are short on timings beyond the screen.

6 roundsHiring call · Coding screen · Second coding round · System design screen · Onsite culture screen · Final round designSee each roundHide
01
Hiring call
A conversation with Decagon, length not reported

Background and fit. One September 2026 candidate lists it as the first step before any technical round.

02
Coding screen
45 minutes, one question, in one April 2026 report

A practical class with follow-ups. Reported problems: a CSAT tracking class with a window of scores, then updates, then a percentile; a shopping cart with undo and redo.

03
Second coding round
Reported in a July 2026 loop

A counting problem: how many legal games of tic-tac-toe exist. An AI Engineer candidate got the same problem and was asked to estimate the order of magnitude first.

Some loops
04
System design screen
Scheduled as one system design round

Decagon-shaped systems: a rule management system where customers configure behaviour rules for AI agents, with drafts, publishing and rollback; security for the Decagon chatbot. The October 2026 candidate was told to focus on data models, APIs and data flow, and said it felt like an object-oriented design round.

05
Onsite culture screen
Conversation

Generic questions, in the September 2026 report: tell me about a project, tell me about a project that went wrong, how do you use AI. That candidate was rejected right after this round, so treat it as a real gate.

06
Final round design
One system design question

Design an agent runtime system. The senior candidate found it broad and vague at the start and wished for more context on scope.

Senior, one report
03

Decagon coding and design questions: what candidates report and how to solve them

Reported problems and how to solve them

Decagon has no timed online assessment in any report. The filter is a live coding screen, reported at 45 minutes with one question, then the design screen. Three coding problems and three design prompts appear across the 2026 reports. Interviewers vary the follow-ups, so practise the approach for each and expect a new requirement halfway through.

6 reported problemsSee them and how to solve themHide
  • A CSAT tracking class: Keep each score with its time in a queue and drop scores that have left the active window when you read. Keep a running sum and count so the average is instant, and round to two decimal places. For updates, store scores by id and adjust the sum. For the percentile, use the fact that scores run from 1 to 5: a count per score answers it without sorting.
  • A shopping cart with undo, redo and versions: Model recipes and discounts as small types and compute totals from one state object. Record every change as an operation with its inverse, or store an immutable snapshot per version. Undo pops from one stack onto another; a new change clears the redo stack. get_version reads the snapshot (or replays operations) for a version number. Decide what checkout does to history and say it out loud.
  • Counting legal tic-tac-toe games: Estimate first: at most 9 factorial, 362,880 move orders, fewer because games stop at a win. Then search depth first, placing X and O in turn, checking for a win after each move, and counting a game when someone wins or the board fills. The answer is 255,168. One candidate solved it and was still rejected, so explain each step as you go.
  • A rule management system for AI agents: Treat it as data modelling. A rule has versions; a customer has a draft; publishing freezes the draft into an immutable configuration with an id; a deployment points an agent at one configuration; rollback moves that pointer back. Write the APIs for each action, then explain how a live agent picks up a new configuration without breaking conversations already running.
  • An agent runtime: Ask what kind of agent and pin down one case. Then the loop: load conversation state and the customer's configuration, call the model, run tool calls with scoped permissions and timeouts, save state, and stop or hand off to a human. Add retries, step and cost limits, a log of every turn, and isolation between customers.
  • Security for an LLM chatbot: Start with the threats: prompt injection, one customer's data reaching another, personal data in answers, tools doing more than the user is allowed. Then controls: tenant isolation, authenticating the end user before account data is loaded, least-privilege tools, filters on input and output, and an audit log with a way to switch a behaviour off fast.

The coding problems reward a small, clean class you can extend while talking. Practise each one in your strongest language against a timer of 45 minutes, and have the interviewer's likely next requirement ready before they ask it.

The rest of the loop depends on your role. Pick it, or paste the posting, and Calibrd predicts the rounds and questions you are likely to get at Decagon.

Predict my Decagon questions →
04

What Decagon screens for

Decagon publishes no interview rubric. What it does publish, on its careers page, in its postings and through its founder, points to the same few things.

  • Speed: its careers page says it values decisive action and speed over prolonged deliberation and planning
  • Being in the office every day, which every posting states
  • Product judgement about customer service: the design rounds are versions of Decagon's own product
  • How you use AI in your own work, asked directly in the 2026 culture screen
  • Clean, extensible code under follow-ups, since each coding problem grows a new requirement
05

Decagon interview questions

Candidate-reported themes

The questions below are the ones candidates reported in 2026, with practice versions where a report names the topic but not the wording.

Rehearse these out loud for Decagon. Free to try.

See what's included →

Behavioural & motivation

  • Why Decagon?
    Cal

    I'm Cal, your AI coach. Say whatever comes to mind: I'll put your answer in order, show what's missing, then score it.

    Scored on Specificity · Connection · Research
    • Your answer is scored on
    • SpecificityNames at least two concrete things about this company or role that could not be copy-pasted to a competitor.
    • ConnectionTies those specifics to their own goals or experience, so it reads as fit rather than flattery.
    • ResearchShows they actually looked: a product, a technical choice, a public post. Generic praise for culture or mission does not count.
    See how a strong answer sounds

    Example · not your answer

    “You're the only team I've found that publishes incident reviews with the real timeline in them, and I've read four now. I've worked somewhere that treated postmortems as a ritual you get through, and I'd rather not do that again.”

    Strong. One row still open.

    • SpecificityOne, not twoThe incident reviews are the only specific here. The rest could be said to any company.
    • ConnectionJoined upTies the observation to how they want to work, not just to admiration.
    • ResearchLooked properlyRead four of them. That is looking, not skimming an About page.

    That comes back in about ten seconds, with what worked, what to fix, and what the question was really after.

    We never store your recording or your answer. How this is handled ↗

  • Tell me about a project you worked on.
  • Tell me about a project that went wrong.
  • How do you use AI in your work?

How Talk it through works

Technical

  • Design a class that records customer satisfaction scores from 1 to 5 and returns the average over the current active window. Then support updates, then a percentile.CSAT tracker: Design a customer satisfaction tracking class: scores from 1 to 5, the average over the current active window, then updates, then a percentile (coding screen, April 2026)Listening for: The window first · Handles updates · Percentile with five values
  • Build a shopping cart for a cooking app with add_recipe, remove_recipe and get_total_discounts. Then add undo, redo, checkout and get_version.Versioned shopping cart: A cart for a cooking app with add_recipe, remove_recipe and get_total_discounts, then checkout and get_version (July 2026); a cart with undo and redo (September 2026)Listening for: Clean data model · Undo, redo and versions · Edge cases
  • On an empty 3x3 board with X moving first, how many distinct legal games of tic-tac-toe are there? Estimate the order of magnitude first.Counting tic-tac-toe games: Count the legal games on an empty 3x3 board with X first (SWE, July 2026); count the distinct game sequences, estimating the order of magnitude first (AI Engineer, July 2026)Listening for: A sensible estimate · Correct search · Checks the answer
  • Design a rule management system that lets customers configure behaviour rules for their AI agents, with drafts, publishing and rollback to a previous deployment.Rules for AI agents: Design a rule management system where customers configure behaviour rules for AI agents: create, update, delete, drafts, publishing and rollback (system design screen, October 2026)Listening for: Versioned data model · Clear APIs · How the agent reads it
  • Design an agent runtime system.Agent runtime: Design an agent runtime system (senior final round, Aced report)Listening for: Scopes it first · The agent loop · Running it safely
  • Design the security architecture for an LLM chatbot that talks to a company's customers.Chatbot security: Design security for the Decagon chatbot; PracHub titles it the security architecture for an LLM chatbot product (system design screen, September 2026)Listening for: Names the threats · Controls at each layer · Audit and response
Your Decagon interview

Don’t walk in hoping. Walk in rehearsed.

Calibrd predicts the questions Decagon is likely to ask for your role, interviews you out loud, and tells you what a hiring manager would hear.

Free · talk any answer through

Ready to go all in? for 90 days · or Unlimited graded answers and 45 mock interviews · no subscription ·
Built by Xiaochen Guo, 500+ interviews on the hiring side

How the questions are predicted

06

Pay

Senior or staff software engineer, San Francisco or New York (posting)$200K–$400K plus equity
Software engineer, platform security, San Francisco (posting)$200K–$330K plus equity
Member of Technical Staff, new grad 2027, San Francisco (posting)$185K plus equity
Software engineer, agents, London (posting)£125K–£180K plus equity
Engineering manager, San Francisco (posting)$280K–$430K plus equity
Agent Deployment Engineer (posting)$175K–$230K in San Francisco and New York; £110K–£150K in London (German-speaking)

These are Decagon's own posted ranges on its Ashby board, read on 14 November 2026. Research engineer roles post the same $200K to $400K as senior engineers, and a Senior Agent Product Manager posts $200K to $285K in San Francisco and New York and £145K to £180K in London. The ranges are wide, so ask early where your level sits in them. Levels.fyi had no usable Decagon data on the same day. Equity is the larger question at a company that says its valuation tripled in under six months: ask the strike price and which round's price the grant uses.

Applying to Decagon? Paste the posting and see whether they’d shortlist you →

Job alerts at Decagon. Not interviewing yet? Calibrd reads the open jobs at Decagon every morning and checks each new one against your resume the way a hiring manager would. Get job alerts at Decagon.

07

How to prepare for a Decagon interview

  1. Practise the shopping cart with undo and redo until it takes you under 30 minutes, then add get_version. It is the most reported Decagon problem.
  2. Write the CSAT tracker with a sliding window, then add updates and a percentile, each as a separate step you can explain.
  3. Count tic-tac-toe games out loud: estimate first, then the depth-first search, then check the answer against your estimate.
  4. For design, practise data models and APIs for Decagon-like systems: agent rules with drafts and rollback, an agent runtime, chatbot security. Scale was not the point in the reported rounds.
  5. Take the culture screen seriously. Have a project, a project that went wrong, and a concrete answer on how you use AI, each in two minutes.
  6. Use a customer service AI agent before the interview and note where it fails. It gives you material for every round.
  7. Plan for the office. Decagon is in-office in every posting, so be ready to say you want to work in San Francisco, New York or London.

This guide covers Decagon's software and AI engineering loop, with what is known of the Agent Deployment Engineer and Agent Product Manager roles. 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 Cloudflare, Snowflake or G42? 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 Decagon mock: spoken answers, coached on the spot. Your first mock is free.

Run a Decagon mock →
08

FAQ & sources

What is Decagon's interview process?

Decagon publishes no interview process: its careers page and all 124 postings read on 14 November 2026 are silent on it. The stages, in order: Hiring call; Coding screen; Second coding round; System design screen; Onsite culture screen; Final round design.

What does Decagon look for in candidates?

Decagon publishes no interview rubric. Speed; Being in the office every day, which every posting states; Product judgement about customer service; How you use AI in your own work, asked directly in the 2026 culture screen.

What questions does Decagon ask in interviews?

The questions below are the ones candidates reported in 2026, with practice versions where a report names the topic but not the wording. Technical themes: CSAT tracker; Versioned shopping cart; Counting tic-tac-toe games; Rules for AI agents; Agent runtime; Chatbot security.

What coding questions does Decagon ask?

Decagon has no timed online assessment in any report. The filter is a live coding screen, reported at 45 minutes with one question, then the design screen. Covered on this page: A CSAT tracking class; A shopping cart with undo, redo and versions; Counting legal tic-tac-toe games; A rule management system for AI agents; An agent runtime; Security for an LLM chatbot.

Is Decagon remote?

Mostly no. Every Decagon posting read on 14 November 2026 says "We're an in-office company", and 99 of its 124 postings that day were in San Francisco, New York or London. Four were listed as remote.

Does Decagon publish its interview process?

No. Its careers page and its 124 postings read on 14 November 2026 do not describe the interviews. The stages and questions on this page come from candidates' own reports from 2026.

How much does Decagon pay?

Senior or staff software engineer, San Francisco or New York (posting): $200K–$400K plus equity; Software engineer, platform security, San Francisco (posting): $200K–$330K plus equity; Member of Technical Staff, new grad 2027, San Francisco (posting): $185K plus equity; Software engineer, agents, London (posting): £125K–£180K plus equity. These are Decagon's own posted ranges on its Ashby board, read on 14 November 2026.

How do I get a job at Decagon?

On 14 November 2026 Decagon had 124 open postings on its Ashby board: 37 in sales, 28 in engineering, 23 on the customer team, 10 deployment strategists, and the rest in finance, the founders' office, people, marketing, product and university hiring. By location, 64 were in San Francisco, 23 in New York and 12 in London, with a few in Australia, Toronto, Austin, Amsterdam and remote. Hiring runs in 6 stages, starting with Hiring call.

How do I get job alerts for Decagon jobs?

Calibrd reads the open jobs at Decagon every morning. Add Decagon to the companies you watch in Calibrd's Job search, with the role you're after, and each new job is checked against your CV: you see only the ones you could be shortlisted for. The first search runs as soon as you start it; after that a free account searches once a week, and with a pass the search runs every morning.

How do I prepare for a Decagon interview?

Practise the shopping cart with undo and redo until it takes you under 30 minutes, then add get_version. It is the most reported Decagon problem. Write the CSAT tracker with a sliding window, then add updates and a percentile, each as a separate step you can explain.

Prep for a real Decagon role

Practise your Decagon interview, out loud.

Paste a real Decagon 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

Two weeks out?

Email me the Decagon prep checklist

The round that decides it, the questions they ask, and what to do before, in one email. Nothing else unless you sign up.

Decagon Interview: Coding, Agent Design, Pay — Calibrd