November 5, 2026

AI-written applications for one senior role: how to screen them

One senior engineering role, hundreds of AI-written applications. How to screen them in a week without rejecting good people, and when to stop posting.

Guide

Recruiting

Don't read the pile. Put the knockouts in the post, send every applicant a 20 to 30 minute written work sample about your real problem before you look at a CV, and check identity only on finalists. Run that gate for two weeks. In parallel, work referrals, direct outreach or a vetted shortlist, because the managers quoted below say inbound produces few hires.

Do you have to read all of them?

No. The Pragmatic Engineer newsletter reports a Swiss startup that closed intake at 600 applications after two days for one senior frontend role. DuckBill Group's CEO told it the company gets about 1,000 inbound applications a day, and "maybe two" are relevant. Engineering managers quoted in the newsletter put the share of inbound applicants who qualify at 10% or less.

A one-person hiring team can't read its way through that, so design the process around the volume. For the Ashby and Greenhouse data behind the problem, see the Ashby and Greenhouse application data.

How do you screen hundreds of applications in two weeks?

1. Cap intake and set an hours budget before you open the inbox

The Swiss startup closed at 600. Do the same, but pick a number of hours, not a number of CVs. Say your week has ten to fifteen hours to give this role (set your own). Everything below has to fit inside that, and when the budget is spent the role stops taking new reads.

2. Put the knockouts in the post, plus a one-question form

State visa or right to work, location and time zone, working language and rate range in the post, and ask for a yes or no on each before anything else. It looks like a small change. It's the biggest cut you can make.

Here's why, using the Swiss startup's reported outcome. Of its 600 applications, about half were rejected for no visa sponsorship and about 30% for non-tech reasons such as weak spoken English or unwillingness to relocate. That left roughly 120 moving forward.

Illustration, not the startup's method: if the post had stated visa and language up front, up to about 480 of those 600 never need a read. Send the work sample to the remaining 120 and read each answer in five minutes (your number will differ). That's about ten hours, which fits the budget above. You read CVs only for the people who pass the gate.

3. Send a 20 to 30 minute written work sample before reading CVs

This is a short pre-CV task, not a take-home. Ask about your real problem, in a form someone can answer in writing in under half an hour. For example: "Here is a failing endpoint and its log. What do you check first, in what order, and why? What would you ask the team?" Write yours from a problem your team actually had.

The sample filters out people with no experience of your kind of system, whatever tool they used to write the answer. Someone who has debugged one answers from memory, and someone who hasn't struggles to fake it.

Reactions to this kind of ask are split. On a Hacker News thread about fake applicants, one commenter claimed a roughly 30-minute pre-application task weeds out 95%, and another called a 30-minute ask insulting. Both are worth heeding, so our rules are:

  • Keep it under half an hour, and say so in the post.
  • Reply to everyone who sends one, even with a two-line no.
  • Offer a 20-minute call as the alternative, so a strong engineer who writes badly can still show how they think.

The longer take-home and live stages come after this gate. Designing the technical screening stages that follow covers them.

4. Score blind against a rubric you wrote first

Write the rubric before the first answer arrives, with three or four items: does it name the likely failure mode, does it order checks by cost, does it say what they'd ask the team, does it state a trade-off. Strip names from the answers. Have two reviewers score the first 20, compare where you disagree and fix the rubric, then score the rest alone.

A fixed rubric also keeps each answer to a few minutes, so the review stays inside your hours budget.

5. Verify identity on finalists only, before the first live call

Veremark's principle is to ask for more evidence as the cost of a wrong decision rises. For a senior remote role that cost is high, so check identity once you have a short list, not on the pile. Fortune reports that many company leaders, Greenhouse among them, now ask for in-person final interviews, so for a senior hire you can make the last round in person, or on a live video call with the checks below. Checking a finalist is a real person has the method. Doing it on the whole pile would spend the hours you just saved.

How do you avoid rejecting good people?

Don't reject on "sounds like AI"

An AI flag can't tell fabricated experience from experience an honest candidate polished with a tool. Why an AI screen is a poor way to reject people goes into the automated side.

Don't filter on keywords

A UK head of engineering told the Pragmatic Engineer that candidates list terms "like RAG, evals, inference" with little behind them. A keyword filter rewards exactly those people, and drops engineers who describe real work in plain words.

Allow AI in the work sample

Let candidates use whatever tools they'd use on the job. Judge the work, not the prose: score the reasoning, so a polished answer with no trade-off in it scores low and a plain one with a sharp trade-off scores high. To see whether your gate rejects people you'd want, have a second reviewer blind-score ten of the rejected answers and count how many you'd have advanced.

Should you stop posting the role publicly?

Not necessarily, but treat the post as one channel to measure. Log your hours per qualified finalist for each channel for two weeks.

Keep the post, behind the gate

The gate above makes the post workable. The catch is that it still costs review hours every week you leave it open.

Referrals and your network

DuckBill's CEO said all recent hires came through the network or LinkedIn outreach. The catch is a narrow pool and a bias toward people who look like you.

Direct outreach, yours or a sourcer's

A hiring manager at an Indian startup turned off inbound postings and uses LinkedIn's recruiter product for outbound only. You choose who you talk to, instead of sorting who turned up. The catch is your time, or a sourcer's fee.

A pre-vetted shortlist

HighCircl's senior engineers run four vetting stages: background and experience verification, a communication and product-thinking assessment, a take-home project mirroring real work, and a live technical session on architectural reasoning. About 10% of applicants pass. You get a shortlist of 3-5 candidates in 72 hours. The catch is that it's a vendor, so weigh it against the other three on your own numbers.

The decision rule

Measure for two weeks. If the work-sample gate leaves you fewer than three finalists worth a call, or your hours go over budget, shift your time to the other channels. If the gate is producing finalists inside the budget, keep the post.

What this means for the senior seat you have open

You're deciding where the next ten hours of your week go: gate, channel or shortlist. Our recommendation is to run the gate for two weeks with a written hours cap, start referrals and outreach the same day, and treat the shortlist as the option you buy when your own numbers say inbound won't fill the seat in time.

You're the scarce reviewer. A channel that eats your hours is expensive however cheap it looks. If the role is still empty because too few qualified people applied rather than too many, when the problem is too few qualified applicants, not too many is the better starting point. For the whole vetting loop after the pile is cut, read how to vet software developers.

FAQ

How many applications is normal for a senior engineering role?

Managers quoted by the Pragmatic Engineer newsletter report 600 in two days for a senior frontend role in Switzerland, and 800 resumes over three months for one software role at a Seattle Series B company. Plan for hundreds, and expect 10% or fewer to qualify.

Can you tell if an application is AI-written?

Often you can suspect it, and managers report CVs "clearly tailored by AI", and the newsletter's author calls cover letters AI-generated and dull. You can't prove it, and it isn't a reason to reject: polished and fabricated are different things. Test what the person can do with a short work sample.

Should you close the role to new applicants?

Cap it, don't necessarily close it. A Swiss startup closed intake at 600, and a Canadian engineering director said their small company "gave up on inbound hiring". Our rule: keep the post behind the gate for two weeks, and move your hours elsewhere if it yields fewer than three finalists worth a call.

Do work samples put off senior candidates?

Some, if they're long. Commenters on Hacker News split between calling a 30-minute task a good filter and calling it insulting. Keep it under half an hour, reply to everyone, and offer a 20-minute call as an alternative.

Sources

Share this article

Author Image

HighCircl Editorial Team

The HighCircl editorial team writes about hiring software engineers, nearshore development, and engineering team building. Our articles draw on direct experience sourcing and placing senior developers across Poland, Hungary, Slovakia, Serbia, Slovenia, Romania, and Spain — and on candid conversations with the CTOs and engineering leads who hire them.

HighCircl is a nearshore engineering network that delivers matched candidate shortlists in 72 hours. Every piece of content we publish is informed by real engagement data: actual developer rates, real hiring timelines, and what separates engineering teams that scale cleanly from those that stall.

Take Me to the Experts

Access our network of industry-leading software engineers.

Start Now