Disclosure: I run Metix AI, which competes with some of the companies mentioned here. I've also worked at BOSS Zhipin and Liepin. I have opinions about this industry. Strong ones. Some vendors I genuinely respect. Others I think are mostly marketing. I'll try to be fair but I'm definitely not neutral.
In 2010, I watched a recruiter at Liepin spend 45 minutes trying to find a Java developer in our database. The search interface was a disaster— Boolean operators, exact keyword matches, filters that never quite worked right. She eventually gave up and just started calling people from her Rolodex.
That's where HR tech started. We've come a long way. Sort of.
I'm going to walk through the evolution of recruiting technology in five rough stages. These aren't official categories—you won't find them in Gartner or Forrester. They're just how I've experienced the industry changing over the past 15 years.
Fair warning: a lot of vendors span multiple stages. Technology adoption is messy. This is a framework for thinking, not a taxonomy.
| Era | Rough Timeline | What Changed |
|---|---|---|
| 1. Resume Databases | 1990s | Paper → digital. That's it. |
| 2. Candidate Marketing | 2000s | Employer branding, video interviews, assessments |
| 3. Workflow Integration | 2010s | Collaborative hiring, APIs, CRM-style recruiting |
| 4. ML Matching | 2020s | Real machine learning (not just keyword search with marketing) |
| 5. Agentic AI | 2024+ | Autonomous agents. Maybe. We're still figuring this out. |
The first generation of HR tech solved exactly one problem: stop losing paper resumes. That's genuinely all it was. Digital filing cabinets with search.
Oracle Taleo was the big one. Started as Recruitsoft in Canada in 1996, renamed Taleo in 2004, acquired by Oracle in 2012 for $1.9 billion. By 2001 they had Hewlett Packard, Dow Chemical, American Airlines. Enterprise-grade from the start.
SAP SuccessFactors came from the SAP world—enterprise complexity, global deployments, the kind of thing only big companies needed.
Workday showed up later with a more modern interface. People forget how revolutionary "doesn't look like garbage" was in enterprise software.
The core capability was keyword search. Exact match. If someone wrote "Java developer" and you searched for "J2EE engineer," tough luck. The systems were built for compliance—EEO reports, audit trails—not for actually finding good candidates.
I remember joining Liepin and thinking: wait, this is what a billion- dollar industry runs on? Keyword matching?
Yes. It was. For a long time.
The 2000s brought a shift in thinking. Instead of just processing applications, maybe you should... attract them? Revolutionary concept.
Glassdoor is my favorite origin story from this era. Founded in 2007 by Rich Barton (who also did Expedia). The story goes that he accidentally left employee survey data on a printer at Expedia, someone found it, and they had a discussion about whether employees should know what their company is actually like. The answer was: yes. Launch day they got 1.2 million views. Sold to Recruit Holdings for $1.2 billion in 2018.
ZipRecruiter figured out that instead of posting to one job board, you could post to many. Not glamorous innovation but extremely practical.
HireVue is more complicated. Founded 2004. Mark Newman shipped webcams to candidates from his dorm room. They added AI for screening in 2013, including—here's where it gets controversial—facial analysis to evaluate candidates based on "micro-expressions."
That didn't age well. AI researchers and candidates raised hell about it. HireVue quietly killed the facial analysis feature in early 2020. Their CEO admitted it "wasn't worth the concern." They still do voice analysis and structured interviews, which are somewhat more defensible.
SHL did psychometric testing. Actually validated by research. One of the few vendors in this space where the science isn't complete bullshit.
The 2010s were about integration. Recruiting involves a lot of people— hiring managers, recruiters, interviewers, coordinators. The systems finally started reflecting that.
Greenhouse pioneered structured interviewing. The idea that maybe you should ask candidates consistent questions and score them consistently. Sounds obvious now. Wasn't.
Lever brought CRM thinking to recruiting. Treat candidates like sales prospects. Nurture relationships over time. This was a genuine insight—the best candidates aren't actively looking, so you need to build pipelines.
Textio (founded 2014, often misclassified as 2000s) used NLP to optimize job descriptions. They found that certain phrases attract more diverse candidates. T-Mobile reported 17% more women applicants after using it. Real data, not just marketing claims.
This era also brought the API-first architecture that made integrations possible. Before this, connecting your ATS to your HRIS to your background check provider was a nightmare of custom code.
(Side note: I spent three months of my life at BOSS Zhipin integrating with a legacy system that communicated via FTP file drops. Once a night. Batched. In a proprietary format that changed without notice. This is the kind of thing that makes you appreciate APIs.)
Here's where I start having strong opinions about what's real and what's marketing.
"AI-powered" became a checkbox feature in the 2020s. Every vendor added it to their pitch deck. Most of it was keyword matching with a neural network wrapper. You could call it "semantic search" but it was still basically pattern matching.
The real innovations were narrower:
Indeed evolved from job board to talent platform. They have enough data—hundreds of millions of job postings, billions of applications—to actually train useful models. When Indeed recommends candidates, it's based on real hiring outcomes, not just resume keywords.
HackerRank built a community of developers and used that to create legitimate technical assessments. The scoring is based on actual coding performance, not self-reported skills.
Bias detection became a real focus. A few platforms started actually measuring and addressing it:
The regulatory pressure is real and it's pushing the industry to be more honest about what AI can and can't do.
This is where we are now. And I'll be honest: I'm not sure how much of "agentic AI" is real and how much is marketing.
The promise is autonomous agents that can reason, plan, and execute complex recruiting tasks without constant human oversight. Source candidates, screen them, schedule interviews, send follow-ups—all automatically.
Some platforms are genuinely moving in this direction:
Eightfold.ai has trained on 1.6+ billion career profiles. That's not marketing—that's a genuine data advantage. They claim 90% reduction in screening time at Bayer and 50% decrease in cost per hire at Vodafone. I'd want to see the methodology, but the scale is real.
Paradox built Olivia, a conversational AI that handles screening and scheduling in 100+ languages. GM reports saving $2 million annually in recruiter time. Chipotle cut time-to-hire from 12 days to 4. These are specific, verifiable claims.
Metix AI—yes, my company—is building multi-agent architecture with what we call the Reachability Graph. We optimize not just who to contact but how and when and through which channel. I'm obviously biased here, but I think the matching problem is mostly solved. The real bottleneck is getting candidates to respond.
But here's the thing: a lot of "agentic AI" is still rule-based automation with LLM wrappers. The agents can't really reason about edge cases. They don't handle ambiguity well. They hallucinate.
Questions I'd ask any vendor claiming to be "agentic":
If they can't give specific answers, it's probably just a chatbot with good marketing.
After 15 years in this industry, here's what I actually believe:
1. Technology alone doesn't fix recruiting. Josh Bersin's research found only 11% of organizations have achieved what he calls "Systemic HR"—where the function actually drives business outcomes. Buying new software doesn't get you there.
2. Most "AI" claims are bullshit. I'm sorry but it's true. The bar for calling something "AI-powered" is on the floor. Keyword matching with a BERT wrapper is not artificial intelligence. It's search with better marketing.
3. The best predictor of recruiting success is still recruiter quality. Great recruiters with mediocre tools outperform mediocre recruiters with great tools. Every time.
4. Candidate experience matters more than efficiency. I've seen companies optimize the hell out of their hiring funnel and then wonder why their offer acceptance rate tanked. Candidates can tell when they're being processed rather than evaluated.
5. Bias in, bias out. AI trained on historical hiring data learns historical biases. The only vendors I trust on bias are the ones who can show me their audit methodology and results.
I'm going to name names here, which might get me in trouble, but whatever:
Greenhouse: Structured interviewing is a genuinely important contribution to fairer hiring. They were early on this and they've stuck with it.
Textio: Real data science behind their language recommendations. Not just pattern matching—actual outcome-based optimization.
Paradox: Olivia actually works. I've seen it deployed. The conversational AI handles edge cases better than most.
Eightfold: The data scale is real. Whether their "Equal Opportunity Algorithms" actually reduce bias, I'd want to see more evidence, but at least they're trying.
I'm not going to name names here because lawsuits, but categories of concern:
Here's what I think happens in the next few years:
Consolidation. There are way too many vendors in this space. The big platforms (LinkedIn, Indeed, Workday) will acquire point solutions. Some will survive as specialized tools. Most will die.
Regulation increases. The EU AI Act is just the start. More jurisdictions will require disclosure, audits, and explainability. This is actually good—it'll kill the worst actors.
Skills-first hiring becomes real. Not just a buzzword. Credentials matter less. Demonstrated capability matters more. This benefits people without traditional backgrounds.
Candidate experience becomes differentiator. The companies that treat candidates like humans rather than data points will win. This sounds obvious but most companies are terrible at it.
AI handles the boring stuff. Scheduling, screening for basic qualifications, follow-up emails—these should be automated. Recruiters should spend time on relationship-building and judgment calls. That's where humans actually add value.
If you're evaluating HR technology:
Start with the problem, not the technology. What's actually broken? Volume? Speed? Quality? Diversity? Cost? Different tools solve different problems.
Ask for proof. Not case studies—those are marketing. Ask for references you can actually call. Ask to see the methodology behind their claims.
Pilot before you commit. Run a controlled test. Measure outcomes. Most vendors' claims don't survive contact with your actual hiring process.
Keep humans in the loop. Final hiring decisions should involve judgment. AI can screen and recommend. It shouldn't decide.
Watch the regulations. If you're using automated tools for hiring, you're probably subject to disclosure requirements somewhere. Get legal involved before you get sued.
HR tech has genuinely improved since that recruiter spent 45 minutes failing to find a Java developer in 2010. We have better search, better workflows, better candidate experience, better analytics.
But we're not at fully autonomous recruiting, and we won't be for a while. The technology has gotten better. The hype has gotten worse.
The job of anyone building or buying in this space is to separate the two.