The interview you cannot cheat,because using AI is the test.

Candidates solve a real ticket with an AI coding agent. Assay records every prompt and every diff, then interviews them on the work they just did.

Two more questions, then you are in.

The problem

Why the interview broke

The tools changed overnight. The interview stayed in 2012.

2×

Rise in candidate cheating (H2 2025)

Rise in reported candidate cheating over the second half of 2025. Overlay tools now render below the layer screen sharing captures, feeding candidates answers in real time.

91%

Of engineers use AI coding agents at work

Most technical screens still ask candidates to invert a binary tree alone, with no internet and no tools, on a sterile virtual whiteboard. It tests what they do not do at work.

Applicants per open role in India have more than doubled since 2022. Companies are not short of candidates. They are short of a way to tell them apart.

The difference

Everyone else tries to catch AI. Assay hands it over.

Detection is an arms race against software that draws below the layer screen sharing captures. Assay does not enter it.

A usual technical screen

The tools

  • AI is banned, then policed with proctoring.
  • Overlay tools render below the layer screen sharing captures.

The test

  • An algorithm, alone, from memory.
  • Nothing like the work the job actually involves.

The cost

  • A senior engineer in the room for every first round.
  • Most of those hours go to people who were never going to make it.

The result

  • A number you are asked to trust.

An assay

The tools

  • The candidate is handed an AI coding agent and told to use it.
  • Every prompt, diff and test run is recorded in order.

The test

  • A real ticket in a seeded repository.
  • Questions generated from the diff they wrote minutes earlier.
  • Nothing to buy: those questions existed nowhere until they wrote it.

The cost

  • Nobody in the room. The interrogation runs itself.
  • Your engineers meet the shortlist, not the queue.

The result

  • Every score opens onto the quote, timestamp or diff behind it.
  • Assay recommends. A person decides.

The session

How a session works

  1. A ticket card reading #3247, fix data sync error, tagged as a bug, beside a repository mark.

    Real ticket

    A seeded repository and a real ticket. Sometimes a bug, sometimes a spec with a deliberate ambiguity in it.

  2. An editor window with an AI agent panel offering plan, edit, run and test actions.

    Agent-enabled work

    The candidate uses an AI coding agent. Every prompt, diff and test run is recorded in order.

  3. A spoken question, why did you choose this approach, beside a microphone and a reply bubble.

    The interrogation

    Assay asks about their own diff, by voice, seconds after they finish. Not general questions. Their questions.

  4. A session report card scoring correctness, reasoning, communication and an overall figure.

    The report

    Every score expands to the exact line, timestamp or diff that produced it.

The report

Every score points at the moment that produced it.

Assay never hands you a number and asks you to trust it. Open any score and you get the exact answer, timestamp or diff hunk behind it. The report is built to be argued with.

An Assay report for anonymised session AS-114. A Strong Hire recommendation with its reasoning sits at the top, followed by four scores out of 100 for technical skill, problem solving, communication and role fit. The interview transcript with timestamps runs alongside, and an evidence panel links to timestamps, direct quotes, reasoning and diff hunks.
  1. 1

    The overall recommendation

    One line, with the reasoning attached. Never a number on its own.

  2. 2

    Scored competencies

    Each one out of 100, and each one opens onto what earned it.

  3. 3

    The evidence behind every score

    Timestamps, direct quotes, reasoning and diff hunks, one click from any score.

See this report on your own candidates.

We are onboarding a small number of teams to run real candidates.

Join the waitlist

The limits

What Assay does not do

It recommends, and a person decides.

Who it is for

Built for teams without a spare senior engineer.

For companies where a first‑round technical screen costs a senior engineer an hour per candidate, and most of those hours go to people who were never going to make it.

Engineers hired a year

20500

This year

Google, Meta and Canva all moved to AI‑assisted interviews. All of their versions need a senior engineer in the room.

This one does not.

Backed byApprify Technologies, built oninfrastructure from Sentry, Deepgram, Sarvam AI, Neon and Anthropic.

  • Apprify Technologies — Backer
  • Sentry — Error monitoring
  • Deepgram — Speech recognition
  • Sarvam AI — Indian language models
  • Neon — Serverless Postgres
  • Anthropic — Model provider

The waitlist

Put your team at the front of the line

We are onboarding a small number of teams to run real candidates before Assay opens to everyone. Tell us roughly how much you hire and we will be in touch.

First

Access before Assay opens publicly

Then

A full session and its report, walked through with you

Built for

Teams hiring 20 to 500 engineers a year

We email you about Assay access and nothing else. See the privacy note.

Onboarding a small number of teams now.

Join the waitlist