Applications are up 239%, and 75% fewer applications reach the hire stage. Those are the headline figures that open the episode. Daniel Chait's own measures are narrower: applications per job are up almost two and a half times since 2022, and with recruiting teams flat or smaller, one recruiter may be looking at four or five times the volume of a couple of years ago, then ask an off-the-shelf model which 20 or 30 of a thousand to call. One researched application that says something true about the role still carries signal, and an AI screen helps when the company explains it first.
Daniel Chait is CEO and cofounder of Greenhouse, the hiring platform used by more than 7,500 companies. Chain of Thought episode 58, which this draws on, was released in May 2026.
What is the AI hiring doom loop?
Candidates use AI to find more jobs and to tailor every resume to the posting. Recruiters filter that pile with off-the-shelf tools and call a short list. Getting seen gets harder, so candidates send more, and companies filter harder. Chait calls that cycle the doom loop.
Each move is understandable for the person making it. Together they drain information out of the process.
"Ultimately, what it becomes is kind of a signal problem. It's like a matching problem"
Daniel Chait, CEO and cofounder at Greenhouse, on Chain of Thought ep 58
The AI Hiring Doom Loop: Appli… - Chain of Thought | AI Agents, Infrastructure & Engineering - Apple Podcasts
Podcast Episode · Chain of Thought | AI Agents, Infrastructure & Engineering · May 6 · 57m
podcasts.apple.com AI customizes every application to the posting, so resumes start to look alike. He describes a thousand of them, similar enough that the signal gets thinner, and a seeker whose own file has been made to blend into that pile.
"It's like you're sending in … a needle to an ever growing haystack, and you're making your needle look more like hay."
Daniel Chait, CEO and cofounder at Greenhouse, on Chain of Thought ep 58
More postings do not clear this. Once the pile is large and uniform, the number of open jobs stops deciding who gets a conversation.
Why are software engineers the heaviest auto-appliers?
On Greenhouse's data, Chait says, software engineers use automation the most and send the most applications. He does not see them as the bad guys. They are using tools that make mass applying easy, and they want a job.
Chait spent ten years hiring programmers before Greenhouse, back when an engineer who answered a recruiter was a rare event. He now describes companies awash in candidates, with thousands on a job and a harder match than the old shortage.
The work moved on a separate track. Software engineering, he says, is already radically different than it was two or three years ago, and than it was sixty days ago. Some announced cuts name AI. He thinks a lot of that is overblown: cuts that maybe were going to happen anyway, after over-hiring, with a better story for the stock market, and he does not think a lot of that represents real productivity gains yet. Cheaper software, in his view, can raise appetite for more of it, because many problems stayed unbuilt while the work was expensive.
What should you send instead of an auto-apply?
Chait's advice on automated blasts is that they feel like progress and fail as a search.
"It's sort of a false progress. You feel like, I've applied to so many jobs, you know, but you really haven't. You've just pushed a button and spun the wheel."
Daniel Chait, CEO and cofounder at Greenhouse, on Chain of Thought ep 58
Count an application when you have done the research, written something real about the company and the job, and, if you can, reached someone who works there. Repeat that. He contrasts it with spray and pray.
Dream Job is the same limit inside Greenhouse's tools. On My Greenhouse, a seeker can apply across customer companies, and once a month mark one application as their dream job. The single mark pushes them toward a role they want and think they can get. Hiring teams see that flag with referrals and internal applicants. Chait says 2,000 people have already gotten their jobs through Greenhouse Dream Job, converting about five times higher than every other job seeker who applies to those jobs.
You can copy the limit without the product. Pick one role you would defend out loud, and attach evidence of work you did.
A minimal sketch to adapt, not a drop-in library.
def classify_application(researched, says_something_real, auto_apply): if auto_apply: return "false_progress" if researched and says_something_real: return "high_signal" return "thin" Referrals still produce the highest chance of a hire among the sources he tracks, and there are too few of them. Most people cannot get one. Asking for advice is the route he names: a person asked for thoughts may know a team, or a friend with an opening. Communities are where recruiters already look. He started a .NET programmers group in the early 2000s, hosted 30 or 40 people a month, and knew whom to call when a role opened.
There is a timing edge while the tools are new to almost everyone. Chait's point to people early in the field is that a new technology is new to everyone. You can have about as much experience with it as almost anyone, which is hard to match on a technology that has been out for ten or fifteen years. Build with it, and show it.
When does an AI interview help the candidate?
When the company says an AI will run it, and the session is fit to use. In Greenhouse's recent survey, almost two thirds of people have come across an AI interview, and almost 40% have walked away because the process was janky. Early tools had uneven voice, interrupted at the wrong moments, and felt like a stand-in for a person. Companies were not always transparent about whether a person or an AI would be on the other end.
Explain the step, and that first conversation can widen the funnel. The candidate gets a shot to show skill, rather than hoping to be one of a few human interviews pulled from a thousand resumes. Chait points to patience, hours that fit a job or caregiving, and less penalty for accent.
Greenhouse published its own rules for AI in interviews because candidates could not tell what was allowed, while the job expected those tools. Take-home cases increasingly fail to show the person's own work once models write them, so in many cases assessment moved to live sessions, including sessions that use the tools the role uses. Spelling out the rules is how he tries to see capability, including from people who are uneasy in an interview and strong at the job.
FAQ
What does the 239% / 75% figure mean exactly?
They are the headline figures that open the episode: applications up 239%, and 75% fewer applications reaching the hire stage. Chait's own figure is a separate measure, per job: almost two and a half times as many applications since 2022. Most recruiting teams held steady or shrank, so one recruiter may be looking at four or five times the applications of a couple of years ago.
Are AI layoff headlines why engineering hiring feels stuck?
Chait thinks a lot of those cuts maybe were going to happen anyway, after over-hiring, with AI as the cleaner story for the stock market. He does not think a lot of that represents real productivity gains yet. The pressure he describes in this market is the match: high volume, and resumes tuned until they look the same.
Should a developer use AI in an interview?
The fair rules are still open. Chait says companies want what the candidate thinks, and they also hire people who will use AI on the job. Greenhouse's move was to write down what is allowed. Follow rules like that, and do the portion of the interview a reviewer can connect to you.
Takeaway
- Count progress as one researched application with something specific about the company and the role.
- Drop auto-apply. A total that climbs after a button press is the false progress Chait describes.
- Ask for advice, spend time in a community recruiters already use, and put a public sample of your work where it can be found.
- If a tool is new to the whole field, build with it and show the result. That is a real experience gap you can close.
- Read "already radically different" as a change in the work itself, and read the 239% and the 75% as a matching problem.
The longer explainer, with the episode clips, is on Chain of Thought. It draws on this episode.
Subscribe to the Chain of Thought newsletter for new episodes and write-ups like this one.
Drafted with AI assistance from the episode transcripts.
