Footage: U.S. Air Force · public domain
For hiring teams · Closed beta

Interview people the way they actually work.

If the job is shipping with agents, the interview should be too. Candidates work on their own machine, in their own editor, with their own agents. You get the whole session back... every file, every agent-written line, your notes on the clock. No browser IDE. No webcam theater.

Get your team on the list
candidate@local — zsh
$
Your editor · your terminal · your agentsNo browser IDENo proctoringAI Usage Policy per interview: AI-Native or TraditionalEvery agent-written line marked, via Git-AINever captured: keystrokes · clipboard · screenHeartbeat every 10 s while connectednpx code-maverick@alpha connectClone an exercise from the shared libraryExercise Network: Front End CalculatorExercise Network: API Incident TriageExercise Network: Agentic Workflow ReviewRecorded · SORTIE-01 · landed 39:30
The receipts

You already know it's broken. So does your entire feed.

For twenty years we've assessed engineers with puzzles they'll never solve again, in editors they'll never use again, with the tools they use every day switched off. The receipts write themselves.

V
Voss ×͜×@uncvoss
Open the post on X

Getting ghosted after interview is becoming normal. Six weeks of checking my inbox wondering if I somehow missed the rejection. And the job posting is still up. You do everything they tell you to do and they can't even send a two-line email saying no. They expect applicants…

September 16, 2026
H
HackerRank@hackerrank
Open the post on X

Leetcode is dead. Nobody writes code line by line anymore. Developers are orchestrating AI, debugging its output, catching when it goes wrong. We're building assessments that test fundamentals and AI fluency together. Not just memorized algorithms. Because that's the actual…

February 9, 2026
K
Kyle Wong@ewveggies
Open the post on X

Leetcode interviews for ML engineers/researchers doesn't make any sense. Oftentimes, it just tests how much you memorized Blind 75. At the same time, a lot of proclaimed "better" ML interviews are really not much better... Finding 3 bugs in this transformer, implementing a…

July 8, 2026
S
Sara Hooker@sarahookr
Open the post on X

Hallelujah. Leetcode was essentially just an indicator of whether you were willing to spend 100s of hours practicing problems that had extremely low correlation with actual every day engineering. +1 for replacing with takehomes or live interviews about realistic codebases.

October 11, 2026
z
zach_the_lizard@zach26655
Open the post on X

I've conducted many hundreds of Leetcode style interviews. Early on it was more reasonable, but as grinding happened the questions got more difficult, code was then expected to be perfect, companies started asking more than one question per round, etc. The people I hired in 2016…

October 6, 2026
G
Gergely Orosz@GergelyOrosz
Open the post on X

Here’s the uncomfortable truth: it doesn’t matter that whiteboard coding tests / LC don’t measure a lot of things a dev needs to do day to day. They are the currently known most economical filter to hire *at scale*. I’m talking when you hire 100+ devs per year (or 1,000+)…

October 5, 2024
E
Emma Bostian@EmmaBostian
Open the post on X

Writing a book about the tech interview made me realize how insane they are. I’m currently writing the chapter on binary search trees and I’m exhausted. Why do we still test this for FE devs? (I was asked to fix a broken tree in a FE interview)

April 19, 2020
A
Ankur💻🎧💪@TheAnkurTyagi
Open the post on X

I quit the interview. tech interviews obsess over coding puzzles but often ignore the real engineering puzzle. I decided at this stage of my career not do interviews who ask live coding or any kind of whiteboard approach because it causes me severe anxiety and pressure and I… pic.twitter.com/g0kwa3Z7eO

March 17, 2024
R
Rareș@Raress96_
Open the post on X

Made it through 3 interview rounds. Role got cancelled. Got rejected at first stage of another role. Zero feedback. Made it to final stages of a third. Job put on hold indefinitely. The tech interview process is broken. Do other people actually land jobs in this market? pic.twitter.com/ok32FyxpFY

May 22, 2026
E
EDOSE✨@iam_biglad1
Open the post on X

The interview process is broken when it takes 5 rounds to decide if someone can do a job they’ll learn in 2 weeks.

August 17, 2026

Sound familiar? There's a better pattern below ↓

Watch one

A real interview, start to finish, from the interviewer's seat.

One candidate, one exercise, one hour, replayed in about a minute. Files show up as they're saved. Agent-written lines get marked as they land. Your notes sit on the clock. The scorecard is the tab in the corner. The live replay runs on desktop. This is a frame from it.

The interviewer's cockpit seventeen minutes into a recorded interview: changed files with agent-written shares, the dispatcher.ts diff, comms, the flight log with notes on the clock, and the scorecard tab.
We don't care if they can grind leetcode in a browser IDE. We care if they can think, and build.

Code Maverick is a technical interview platform for teams hiring software engineers. Candidates work on a real task in their own editor with their own AI agents; interviewers watch live, take notes on the clock, and review the full record, including which lines an agent wrote.

For hiring teams

Run the whole loop from one place.

Exercises, invites, the live floor and the debrief. No spreadsheet, no calendar archaeology, no notes app.

Exercise NetworkShared

Never write another interview problem from scratch.

Pick an exercise to arm it

Real tasks from real repos, used by other teams. Pick one, clone it, change what you want.

Schedule

Invite a candidate in one link.

  1. 01Set upPick the exercise, the candidate, the time.
  2. 02InviteOne link. No account, no download page.
  3. 03PreflightThe candidate checks their machine the day before.
  4. 04StartThe clock starts. The brief unlocks.

Pick an exercise, pick a time, send one link. The candidate checks their setup the day before. The clock starts when you say.

Live board

See every interview your team is running. Jump into any of them.

2On deck
1Airborne
1Debrief
  • SORTIE-01Recorded CandidateAirborneWatch
  • SORTIE-02Front End Calculator · 14:00On deck
  • SORTIE-03API Incident Triage · 15:30On deck

Everyone on your team, one screen. Who is up next, who is in the room, who is waiting on a call.

Make the call

Make the call while it's fresh, with the record in front of you.

  • SORTIE-01Webhook delivery you can trustEffective
    LandedNotesRubric▸ Call
  • VIPER-07API Incident TriageRubric open
    Landed▸ NotesRubricCall
  • GHOST-04Front End CalculatorAwaiting debrief
    ▸ LandedNotesRubricCall

Every finished interview sits here with your notes, the diff, and the rubric. Nothing to reconstruct from memory.

For candidates

The interview that feels like the job.

Your editor, your tools, your agents. Candidates who are treated like engineers show you how they engineer.

Your own cockpit

Your editor. Your terminal. Your machine.

candidate@local — zsh
$

One command in the terminal you already have open. Every save syncs. Nothing to install, nothing in a browser.

Check your setup the day before

Day-of setup problems, solved the day before.

$ npx code-maverick@alpha doctor

  1. Code Maverick CLI
  2. Backend reachability
  3. Sortie packet config
  4. Watched directory
  5. State root
  6. Git probe
  7. Hidden ledger probe
  8. Git-AI runtime
  9. AI tool hooks

One command on the candidate's machine, the day before. 8 pass, 1 warn, nothing to debug when the clock starts.

Agents on the record

Leverage your agents as you normally would.

+import { createHmac, timingSafeEqual } from "node:crypto";
+import type { Endpoint, EndpointSecret } from "./types";
export const SIGNATURE_HEADER = "X-Webhook-Signature";
export const TIMESTAMP_HEADER = "X-Webhook-Timestamp";
+function validTime(time: number): boolean {
Codex (gpt-6-astra)Prompt record

Turn 4 · Implement requirement 3, signing: HMAC-SHA256 over '<timestamp>.<body>' with the current secret, X-Webhook-Timestamp and X-Webhook-Signature headers, and rotation where a previous secret verifies until its expiresAt. Test rotation during the grace window and after it. Keep bun test green. Commit.

42 lines from this prompt

Every line an agent wrote is marked, with the prompt behind it. Driving agents well is a skill. Now reviewers can see it.

Transparent by design

You see exactly what is captured. Before you start.

Captured
  • File contents, paths and hashes inside the exercise directory
  • Sync events and a heartbeat every 10 seconds
  • Your name, as you type it on the candidate page
  • Which lines an agent wrote, where your tools support Git-AI
  • The prompts behind those lines, where your tools support Git-AI
Never captured
  • Keystrokes or clipboard
  • Screen or webcam
  • Browser history
  • Files outside the exercise
  • Secrets: .env, *.pem, *.key
Inspect it: npx code-maverick@alpha --help

The sync client runs when you run it and stops when the interview ends. What it captures is public, and the client is on npm.

What every candidate is owed

The Candidates' Bill of Rights

Ten rights for anyone sitting a technical interview, anywhere. Under each one is what Code Maverick does about it, marked by what's shipped and what's still vision.

For candidates · Be one of the first to sign

Your email is your signature. Nothing else is recorded.

  1. Your own cockpit.

    Your editor, your terminal, your machine. The environment you ship from every day is the environment you interview in.

    TodayWhat Code Maverick does

    No browser IDE, ever. You connect from your own terminal with npx code-maverick@alpha connect.

  2. Your instruments.

    The tools you use every day, agents included. Banning them tests a job that doesn't exist.

    TodayWhat Code Maverick does

    The brief states the AI Usage Policy up front. AI-Native interviews clear your tools.

  3. Attribution, not surveillance.

    No webcam, no keylogger, no screen capture. Record what was written and by what, nothing else.

    TodayWhat Code Maverick does

    Git-AI marks the lines an agent wrote, and the capture contract lists everything recorded.

  4. Real work.

    A task from a real repo with real constraints, not a puzzle you'll never solve again.

    TodayWhat Code Maverick does

    Exercises are real tasks in a real repo you run locally.

  5. The rules before you fly.

    What you're building, what's allowed and what's being assessed, in writing, before the clock starts.

    TodayWhat Code Maverick does

    The brief and the AI Usage Policy are in the invite.

  6. A preflight.

    Check your setup before the day. No surprises at the start of the clock.

    TodayWhat Code Maverick does

    Setup Preflight and the doctor command run before the interview.

  7. Your time.

    A published duration and a scheduled start. The clock is the clock.

    TodayWhat Code Maverick does

    Every interview has both.

  8. Equal footing.

    Every candidate for the role gets the same exercise, the same loadout and the same clock.

    TodayWhat Code Maverick does

    One schedule, one configured exercise, every candidate.

  9. The record.

    The same evidence your reviewers see, not a memory of a screen share.

    VisionWhat Code Maverick does

    On by default. A team can opt out, and that choice is visible.

  10. No ghosting... seriously!

    Candidates go through the ringer for you. The least they get back is an answer. As an industry we're better than this.

    VisionWhat Code Maverick does

    Zero tolerance. Every team on Code Maverick commits to a response, we track who honors it, and teams that ghost face penalties. We expect better.

Questions

The ones hiring teams ask first.

Short answers. The long ones are in the Bill of Rights and the capture contract above.

Can candidates use AI during a Code Maverick interview?
Yes, when the interview's AI Usage Policy says so. AI-Native interviews clear the candidate's own agents and tools; Traditional ones don't. Either way the policy is stated in the brief before the clock starts, and every agent-written line is marked in the record.
What does the candidate need to install?
Nothing in a browser. The candidate runs one command in their own terminal, keeps working in their own editor, and every save syncs to the interviewer. A setup check the day before catches machine problems early.
What does Code Maverick record?
File contents, paths and hashes inside the exercise directory, sync events, a heartbeat, the candidate's name, and, where their tools support Git-AI, which lines an agent wrote and the prompts behind them. Never keystrokes, clipboard, screen, webcam, browser history, files outside the exercise, or secret files (.env, *.pem, *.key).
Is this a replacement for take-home assignments?
It's the take-home done live. The candidate gets a real task from a real repo, a published duration and a scheduled start, works in their own environment, and you watch and take notes instead of guessing at a zip file a week later.
How is this different from HackerRank or CoderPad?
Most interview tools put the candidate in a browser IDE on a puzzle. Code Maverick runs a real engineering task in the candidate's own dev environment, agents allowed and every agent-written line attributed, and shows you how they actually work: files landing, your notes on the clock, and a scorecard at the end.
What does it cost?
Free during the closed beta. Teams get in a few at a time; join the list on this page. The capture client is public on npm (npx code-maverick@alpha --help).
For hiring teams · Closed beta

The list is open.

We're letting teams in a few at a time. Get yours on the list. Free during beta.