Why AI Will Not Replace the Recruiter (and What Your Career Has to Do With It)

Intro: hype, investors, and the unglamorous truth

The last couple of years brought a real AI boom. ChatGPT stunned the world, and startups lined up to promise a hiring revolution: the model would find the perfect candidate, filter out the weak ones, and do it faster than any HR manager.

As the founder of an interview-prep platform, I have talked to hundreds of VCs. The question I heard most often: “Why aren’t you using ML/AI to match candidates? That would make you billionaires!”

I am convinced it is a dead end. Not because the tech is too immature. Because of data. More precisely: the lack of it.

This is not an article about AI writing a cover letter or filling a calendar (it already does that well). The real question: can an algorithm replace a human judging another human?

Spoiler: no. Here is why. While you read, practice what a profile cannot show: a mock with a live engineer, an express quiz, and work that is not on GitHub in pet projects.

Part 1. What do you train AI on — and why it fails

Picture the perfect algorithm: resume in, perfect company out. To train it you need three things:

• Job descriptions.
• Public candidate data (LinkedIn, GitHub).
• Outcomes: who got hired where, and how they actually performed.

It looks like all of that is public. The devil is in the details. Most startups try to build matchers on exactly this “public” pile. That is the mistake.

A hire works only if the candidate hits three bars:

• Good: competent enough for the work.
• Looking: open to offers right now.
• Interested: actually wants this company.

This is the wall. Public data answers none of those questions.

Part 2. The evaluation myth: LinkedIn is guesswork

Recruiters have stared at resumes for decades: school, jobs, skills. Research puts the predictive power of a resume at a coin flip — 50/50.

• Education? Almost no signal.
• Certificates? Often a negative signal.
• Skills? Anyone can type “Python” or “Leadership”.

An algorithm reading LinkedIn does what a mediocre recruiter does: guesses from a pretty list of employers. AI cannot “touch” a candidate’s skills by reading a profile. It is water.

If you still need a resume, do not polish it for a parser. Draft wording in the resume cabinet and check the clarity guide. You will still have to talk out loud.

GitHub is a trap for geeks

“But code doesn’t lie!”

Fair: if someone has public repos, you can judge style and logic.

The problem is GitHub exists for only about 10% of developers. Useful activity (not “Hello World”) is under 1%. Bjarne Stroustrup, the creator of C++, would fail a “coder quality” test scored on GitHub commits.

Most strong engineers at large companies do not ship open source. Their best work sits in private corporate repos. Scoring a candidate on GitHub is looking for a needle in a haystack while most needles sit in other stacks.

What you can show outside is better as a pet project: not another todo app, but something with users, a deploy, and a trail of decisions.

The social graph does not work

The hope: “If a great engineer follows another great engineer, we can draw a talent map.”

We surveyed our users — engineers with about eight years of experience on average. Result: a developer follows only one of their five strongest former colleagues. LinkedIn is a junkyard of random ties. Twitter/X is for memes, not networking.

Bottom line on “Good”: we do not have the data to say if someone is good. AI cannot pull that out of thin air.

Part 3. Do they even want the job — and what do they need?

Suppose we found a genius. Now we need to know if they are looking.

You can watch LinkedIn updates, tenure, whether the company is collapsing. Those are proxies. Many developers do not touch their profile for years, hang out on Reddit (where the model cannot go), and dislike social networks.

The hard part is Interest.

Even if we know the salary band (thanks, new US laws), how do we know they will like the team? The atmosphere? The work?

Companies write about “mission” and “culture” with the same copy-paste words.

Candidates… lie. Not out of malice. They do not know what they want.

A real story: a candidate told me he dreamed of EdTech and hated ads. He passed every interview at our education company — then joined an AdTech startup. Because he liked the team. He liked the chemistry.

That is the whole point.

Part 4. Chemistry you cannot program

Hiring is not matching the tags “Java” and “Spring”. It is finding people who can stand working with each other.

Can you predict whether two people will click?

Can AI score a tech lead’s joke or a Slack voice?

Until we have data on millions of real conversations and emotions, the algorithm is a thin imitation.

Even with ATS employment databases, the best predictor is usually plain linear regression. For messy social forecasts, Princeton research put it bluntly: AI is no better than manual scoring on a few features.

What this means for you as a developer

This is not a reason to fear AI. It is the opposite: your human edge is not going away.

Robots cannot judge code from a resume. For now, the best proof of skill is a live interview. And you have to prep for it.

Chemistry decides. Companies hire people they can stand talking to. Soft skills, explaining your thinking, working with a team — that is your weapon against algorithms.

The market stays human. AI will take the chores (outreach, calendars). The “hire / no hire” call stays with people. That is why you need to present yourself in an interview, not only write pretty code in silence. Practice on a mock interview and close gaps in quizzes.

Closing

AI will be a strong assistant: keyword search, meeting times, first hints for recruiters. It will not replace the live conversation where a candidate’s fate is decided.

Why write this in the IT Career Gym blog? Because our job is to get you through that human filter. We do not believe in resume magic. We believe in practice. Only by running technical interviews with real engineers, again and again, do you show a level LinkedIn cannot see.

Automation is coming, and it removes the boring work. The interesting part — finding talent and building teams — stays with people. Prep for that live part today: create an account and run your first mock on /find/.