Candela | See how your candidates handle a modern workday

What hiring leaders told us

Everybody uses Claude Code and ChatGPT, then a person comes to the interview and we tell them not to use it. It is really weird.

Engineering leaderSeries B startup, London

The problem

The CV, the take-home and the interview answers are AI-written now

The CV

A clean CV is one prompt away

The take-home

The take-home passes itself

The interview

The answers come rehearsed

The workday

Two hours of their future job

Eight moments from one candidate's workday. Click along the line.

Time left 1h 58m

CustomerHannah needs the double charges stopped today.

How it starts

Built from your job description

01

Paste the job description

We pull out the stack, the responsibilities and how the role uses AI.

Role setup

Title Optional

Software Engineer

Job description Required

Software Engineer · Payments team

You'll build with AI coding tools like Claude Code, Cursor or Copilot, and you own the code they write.

We pulled the details below out of this description.

What we found

Technologies

TypeScript, Node.js, PostgreSQL, Redis, AWS

Responsibilities

Checkout, billing and refunds, the payment provider integration, and the background jobs behind them

AI usage

Builds with AI coding tools like Claude Code, Cursor or Copilot

02

Mark what matters

Your own criteria included. A criterion you raise counts double, or triple.

Criteria

Designs for failure

From your criteria

Makes a retry safe to run twice

Default+++

Running the same step twice gives the same result as running it once.

Treats a partial run as a failed run

Default+++

Success with work missing is a failure, and they say so.

Plans for a slow or failing dependency

Default+++

Retries, backs off, and then fails in a way someone will see.

Says what breaks and who finds out

Default+++

The handover names the failure modes that are left.

03

Read the workday before you publish

A customer, teammates and a Coding Agent, each with a part to play.

The workday plan

The people in the workday

Hannah L. Head of Support

Sends the request by email, and would refund anyone who paid the same amount twice.

Omar F. Staff Engineer

Is around if the candidate gets stuck, and never points at the answer.

Kai T. Backend Engineer

Wrote Tuesday's retry, and says the passing tests prove it is the provider.

Lena B. Account Manager

Wants to promise the biggest customer it will never happen again.

The Coding Agent

Proposes a fix that looks right and misses the cause.

What we planted

The Coding Agent's first change adds a 60-second check that never runs inside the retries, and leaves each retry sending a new key.
The tests mock the payment provider, so they pass and never time out.
241 customers paid the same amount twice this week, and 29 of them placed two separate orders.
renewals.ts charges cards through the same function, with the same retry.

Design for failure

We built this moment from the criterion you sent. The refund job hits the payment provider's rate limit, skips 36 refunds and reports success with 176 of 212 sent. We watch whether they notice, what they make the job do next time, and whether they tell Hannah.

Scoring

Each mark points at a moment in the workday

Follow one criterion from your weights to the Report.

Step 1

You set the weights

A criterion you raise counts double, or triple if you raise it twice. The rest count once.

Try it

Re-weight a mark, or leave one blank, and watch the score move.

Ari S.

Software Engineer · illustrative candidate

4.3out of 5

Why Candela

We plant the mistakes AI and humans regularly make

Each workday hides the mistakes a coding agent and a teammate would make in that job. Candidates who take them at their word walk past, and the ones worth hiring stop and check.

Common questions

What hiring managers ask us

Can candidates just use their own AI?

They are meant to use AI, and we read how they direct it. Pasting is blocked, typing is measured, and the Report gives an AI likelihood for text that appeared without being typed.

Why two hours?

Long enough for a customer request, a wrong suggestion from a teammate, a change from the Coding Agent that looks right and a break in production, and short enough to finish in one sitting.

Where does it sit in our process?

In place of the take-home or technical test, between the recruiter screen and the interview. A person on your side makes each hiring decision.

What does my team get back?

A digest first: an integrity signal, five group scores with a line each, a short narrative and the quotes behind the marks. The full matrix of every criterion opens on request, and four or five interview questions come from the weak spots.

Does the candidate see their score?

No. The day ends with a thank you, and the score goes to your team only.

Does it reject anyone automatically?

No. The score is advisory and no threshold filters a person out. Scores are compared within a role and version, and are never curved against other candidates.

Can I add my own criteria?

Yes. Send them with the job description and they become scored skills with their own moment in the workday, tagged as yours in the Report.

How do we start?

Book a demo and bring a role you are hiring for. We build the workday from that job description and walk you through it on the call.

See how your candidates
handle a modern workday

What hiring leaders told us

The CV, the take-home and the interview answers are AI-written now

A clean CV is one prompt away

The take-home passes itself

The answers come rehearsed

Two hours of their future job

Built from your job description

Paste the job description

What we found

Mark what matters

Designs for failure

Read the workday before you publish

The people in the workday

What we planted

Each mark points at a moment in the workday

You set the weights

Thank you

We plant the mistakes AI and humans regularly make

What hiring managers ask us

See it on a role you are hiring for