Chakra Ran 500,000 AI Interviews. Don't Feed It Your Sourced Engineers

If you spent weeks getting a passive staff engineer to reply, sending them to an AI-run first round by default is the quickest way to lose them. That round is now a standard product: HackerRank's Chakra came out of beta on October 5, 2026, after running more than 500,000 interviews, and Ashby is putting its own AI Interviewer into private beta.

What HackerRank launched on October 5

HackerRank made Chakra, its AI interviewer, generally available on October 5, 2026, after a six-month beta that ran more than 500,000 interviews. It runs the technical interview and scores the candidate, and humans make the final call.

The facts:

• Beta customers: Snowflake, Snorkel and Capgemini.
• The pitch: CEO Vivek Ravisankar says Chakra collapses the screen, the take-home and the engineer round into one sitting, and it scores reasoning as well as answers.
• Integrity: HackerRank reports suspicious-activity flags 70 to 80% lower than on its traditional assessments.
• Positioning: HackerRank now says the system scores candidates and humans make the final call.

The integrity number comes from a simple mechanism. Chakra lets candidates use AI openly, so there's no reason to secretly run outside tools in another window. For anyone screening thousands of anonymous applicants, that's real progress.

Ashby's version came out of its Talent Llama acquisition (co-founders Adam Stokar and Ben Chaikin). It's rolling out in private beta, with broader availability planned in the coming months. Ashby CEO Benji Encz described the category carefully: "Tools like AI interviewers are becoming more relevant for certain types of hiring, especially higher-volume roles." Hold on to that qualifier. The rest of this article follows from it.

Why do candidates walk out of AI interviews?

The most common reasons candidates give for walking out are that no human was present and nobody told them how the AI would be used. They aren't objecting to AI as such. Greenhouse's 2026 survey of 2,950 people found that 63% of US candidates had been interviewed by AI in the past 12 months, 38% had already withdrawn from a process because it included an AI interview, and another 12% said they would withdraw if one were required.

38%

The reasons they gave were specific:

The people who stayed didn't do much better. Of the candidates who finished an AI interview, 28% advanced, 13% were formally rejected, and 51% never heard anything back. That follows from how the tool works. A bot makes it nearly free to put more candidates through a round, but the number of humans available to follow up stays the same. Throughput grows and responsiveness doesn't.

Greenhouse's Daniel Chait put it bluntly: "Most AI in hiring today is making a bad system worse: more applications, less signal, and less transparency."

Here's the detail that changes the argument: only 19% of US candidates said they want less AI in hiring. People object to being left alone with an AI and not being told what it's doing.

Why the 38% is a floor for sourced engineers

For sourced engineers, read the 38% walk-away rate as a floor, not a ceiling. Greenhouse surveyed active job seekers, people with an application on the line. A passive engineer you sourced has nothing on the line, so walking away costs them nothing. To be clear, this is an inference. No survey isolates passive candidates.

It isn't "senior people hate AI"

The easy story is that experienced engineers look down on bots. The data says otherwise. An Enhancv survey found that 36% of 18 to 24 year olds and 35% of 25 to 34 year olds had abandoned a process over AI screening, against 21% of people 55 and older. Older candidates were less likely to walk.

What actually drives it is who has invested in whom:

• An inbound applicant found your posting, decided to apply, and expects to clear some hurdles. They pursued you.
• A sourced engineer got a message out of nowhere and gave you a reply as a favor. You pursued them.

An asynchronous bot round shifts the work onto the person you just courted. You asked for their attention, they gave you a little, and you answered with a link to an unattended test.

An inbound applicant chose you. A sourced engineer was chosen by you. The bot round bills the wrong party.

Ashby's own numbers back this up. In an internal test that Recruiting Manager Anna Hall described, candidates could choose between a recruiter screen and an AI interview taken on their own time. Only 36% chose the AI. The other 64% chose the human.

The fraud win doesn't transfer to outbound

Chakra's 70 to 80% drop in cheating flags fixes a problem in anonymous, high-volume pipelines. A sourced engineer was found through verifiable public work, so you already have evidence of identity and skill before the first message goes out. A fraud filter has little to catch there.

Think about what the AI round is for. Inbound pipelines are flooded with polished resumes you can't verify, and at that volume you need a fast, cheat-resistant way to separate real skill from noise. Chakra does that job well.

Outbound works the other way round. You picked the candidate because of things you could already check:

• Merged pull requests in projects other people maintain
• Maintainer or reviewer history, where other engineers trusted their judgment
• Conference talks with a face and a voice

None of this is proof on its own, and a quick identity check is still worth doing. But the identity question that justifies an AI fraud screen has mostly been answered by the time you send the first message. Running the candidate through a fraud filter anyway adds friction and almost no new signal.

How big is the pool you'd be burning?

In Refolk's index of professional profiles, 244,898 people in the US currently hold a Senior, Staff or Principal Software Engineer title, and only 64,317 of them are Staff or Principal. Chakra's beta alone ran about 7.8 interviews for every one of those US Staff and Principal engineers.

7.8x

Staff and Principal engineers make up about 26% of the senior pool. That's the group outbound teams spend the most effort to reach, and it's small enough that every lost candidate counts. In a sample of matched profiles, they clustered in San Francisco and Seattle, at companies like Databricks, Datadog, Starburst and Airbyte. That sample is only 25 profiles, so read it as an illustration, not a ranking.

A tool built to run half a million interviews is aimed at volume. Even Codility, which competes with HackerRank and so has a reason to say this, argues that "the evidence in favor of AI interviews comes from high-volume entry-level hiring." Discount the source if you like. The 1.49x ratio between entry-level and senior pools points the same way regardless.

Finding the specific people within that 64,317 is the expensive part. It's the step Refolk is built for: you describe the engineer you want in plain English and get back people matched on what they've actually shipped. Once you've paid that cost, losing the candidate in the next step is a waste.

Split the funnel: which candidates go to the AI interviewer

Route candidates by how they reached you, not only by level. Inbound applicants can go through an AI interviewer, with disclosure, because the volume and fraud problems are real. Sourced candidates skip it and go straight to a human, because you've already done the screening work.

Two things apply to both paths:

• Disclose everything. New York City now requires independent bias audits for AI hiring tools and requires telling candidates when AI is used in hiring decisions. Failure to disclose was also the second most cited reason candidates withdrew (27%), so do it everywhere, legal requirement or not.
• Keep a human in the loop and say so. HackerRank already frames Chakra as "it scores, humans decide." Tell candidates that directly.

You can also handle edge cases with this rule. A strong inbound senior with no public work history? An AI round with disclosure is reasonable, because you have nothing else to go on. A referral from someone on your team? Treat them like a sourced candidate.

How to screen a sourced engineer without an AI round

Do the technical screening before outreach using public work, then use one human call to test interest and depth. This keeps the evidence you need and drops the step that loses people.

1. Read the work before you write. Pull merged PRs, review threads, issues they opened, and any talks. Look for scope (did they change something load-bearing?) and judgment (how do they respond to review?). Refolk turns this from an afternoon of tab-switching into one query, because you can ask for engineers by the work itself, for example "staff engineers in Seattle with maintainer history who have reviewed more than they've merged."
2. Write outreach that cites a specific contribution. Naming a PR shows you did the screening already and sets up the call as a conversation instead of a test.
3. Make the first call human, and include an engineer. Ask them to walk you through a change they shipped: why that approach, what they'd do differently, what broke. You'll learn more than you would from a generic coding prompt, and it costs the candidate nothing they wouldn't enjoy talking about.
4. If you need a work sample, make it optional, disclosed and AI-allowed. This is where a tool like Chakra can fit, later in the process, once the candidate is invested. Candidates objected to having no human and no disclosure, not to AI itself, and the "AI tools allowed" framing is what drove HackerRank's lower cheating flags anyway.
5. Close every loop within days. 51% of AI-interviewed candidates never heard back. Silence loses any candidate, and a passive one won't follow up to ask.

The order matters. AI rounds put the screening first because the candidate is unknown. With a sourced engineer, the screening already happened when you found them, so the first live conversation should be about whether they want the job.

FAQ

Should I use AI interviews for senior engineers?

For inbound senior applicants, an AI interviewer with clear disclosure and human review is defensible, especially if you're dealing with fraud or volume. For senior engineers you sourced, don't make it the first step. They didn't apply, they have no application on the line, and the most cited reason for walking out in Greenhouse's survey (33%) was a scored interview with no human present. Screen sourced engineers on their public work, then put them on a call with a person.

How much candidate drop-off do AI interviews cause?

Greenhouse's 2026 survey of 2,950 people found that 38% of US candidates had withdrawn from a process because it included an AI interview, and another 12% said they would if required. In Ashby's internal test, only 36% of candidates chose an AI interview over a recruiter screen. Greenhouse surveyed active job seekers, so drop-off among passive, sourced engineers is probably higher, though no survey measures that group directly.

Does HackerRank Chakra actually reduce cheating?

HackerRank reports suspicious-activity flags 70 to 80% lower in Chakra interviews than in its traditional assessments. The likely reason is that Chakra allows AI use openly, so candidates have no reason to hide outside tools. That matters for anonymous inbound pipelines. It matters much less for sourced candidates, whose identity and work you've already checked through public contributions.

Are there legal requirements for using an AI interviewer?

Yes, in some places. New York City requires independent bias audits for AI hiring tools and requires telling candidates when AI is used in hiring decisions. Even where it isn't required, failure to disclose was cited by 27% of candidates who withdrew, so tell every candidate, inbound or outbound, when and how AI is involved and that a human makes the final decision.