AI Hiring Tools Are Quietly Killing Career Switchers

I used to work as a recruiter. Briefly. It wasn’t for me, but it gave me something valuable: a look behind the curtain of how hiring actually works. And what I see happening now with AI screening tools worries me.

Not because AI in hiring is inherently bad. But because the people building these tools are training them on data that says “a good candidate looks like this.” And “this” is a straight line from university to internship to junior role to senior role. If your career looks like that, congratulations. The algorithm loves you.

If your career looks like mine (hospitality, Twitch streaming, a failed recruiter side quest, then a traineeship at 27) the algorithm has already thrown your CV in the bin before a human ever saw it.

The Numbers Are Ugly

This isn’t speculation. 19% of companies that use AI in hiring admit their tools have screened out qualified candidates. That’s the number they admit to. The real number is higher.

Only 26% of companies require a human to review every AI rejection. The other 74% let the algorithm decide autonomously who gets a chance and who doesn’t.

Workday’s AI screening tools rejected one candidate from over 100 roles despite him being qualified. One hundred applications, zero interviews, because the algorithm decided he wasn’t a fit before any human could disagree.

What Gets You Filtered Out

AI screening tools look for patterns. They’re trained on historical hiring data: who got hired, who performed well, who stayed. That sounds reasonable until you realize what it actually means:

Employment gaps. Took a year off to start a business? Spent four years streaming? Cared for a family member? The algorithm sees a gap where it expects continuous employment. That’s a penalty.

Non-traditional job titles. “Twitch Streamer” doesn’t map to any keyword the algorithm is looking for. “Sandwich shop owner” doesn’t either. The skills transfer. Crisis management, customer communication, performing under pressure. But the algorithm doesn’t understand skill transfer. It understands keywords.

Missing credentials. No degree? Filtered. Wrong degree? Filtered. Degree from a university the algorithm hasn’t seen in its training data? Also filtered.

Non-linear progression. The algorithm expects: junior, medior, senior, lead. In that order. If you went from bartender to trainee to junior sysadmin to DevOps engineer, that pattern doesn’t match. It doesn’t matter that the trajectory shows ambition and growth. It doesn’t look like the pattern.

The Amplification Problem

Here’s what makes this worse: research from the University of Washington found that human recruiters who work with biased AI tools mirror the AI’s discrimination up to 90% of the time. The AI doesn’t just filter candidates. It trains the humans using it to think the same way.

So even when a human does review the remaining candidates, their judgment is already shaped by the AI’s bias. The tool that was supposed to remove human bias is instead creating a feedback loop that makes bias systematic.

Amazon scrapped their AI hiring tool after discovering it penalized resumes containing the word “women.” That’s the one that made the news. How many other biases are running in production right now that nobody has caught?

Why This Matters For IT

The tech industry constantly complains about the talent shortage. We can’t find enough engineers. We can’t fill these roles. The pipeline is broken.

Meanwhile, the same industry is deploying AI tools that automatically reject candidates who took a non-traditional path into tech. The career switchers. The self-taught engineers. The people who learned by building, not by studying. The people who bring diverse perspectives because they’ve actually done something other than write code their entire life.

These are often the most driven candidates. You don’t switch careers at 27 because you’re lazy. You don’t spend your evenings learning Azure after working a full day because you lack motivation. You don’t survive a traineeship with a 4-hour daily commute because you’re not committed.

But none of that shows up in the pattern the algorithm is looking for.

What I’d Tell Companies

Stop letting AI make the final rejection decision. Use it to surface candidates, not to eliminate them. There’s a massive difference between “here are the top 50 matches” and “here are the 200 people we rejected without looking.”

The Colorado AI Act, effective June 2026, will require companies to use reasonable care to prevent algorithmic discrimination in hiring. That’s a start. But legislation moves slower than deployment.

What I’d Tell Career Switchers

The system is working against you. That’s not paranoia, it’s documented fact. Here’s what you can do about it:

Optimize for the algorithm even though it’s stupid. Put the right keywords in your CV. List your certifications prominently. Use standard job titles even if your actual title was different. You shouldn’t have to do this, but you do.

Go around the algorithm. Apply directly to hiring managers on LinkedIn. Ask for referrals. Go to meetups. The best jobs I’ve seen filled were never posted on a job board where an AI could screen them.

Build proof that bypasses the filter. A portfolio site, public repos, blog posts. These are things a human can look at that an algorithm can’t score. When someone Googles your name and finds real work instead of just a LinkedIn profile, that matters more than any keyword match.

Don’t give up. Being rejected by an algorithm doesn’t mean you’re not qualified. It means a pattern-matching tool didn’t recognize your pattern. That’s a limitation of the tool, not a limitation of you.

The Irony

I work with AI every day. I build AI agents. I think AI is genuinely useful for a lot of things. But using AI to decide who gets a chance at a career and who doesn’t, based on whether their life story matches a template, is one of the worst applications of this technology.

The best engineer on my team might be someone whose CV would never make it past an AI screening tool. A career switcher, a self-taught developer, someone who started late but works harder than everyone who started early.

We’ll never know, because the algorithm already said no.