I Worked on an AI Recruiter in 2024. Two Years Later, Every Finding Held.

The wedge that works in AI recruiting isn't where most people are building. Here's what product discovery in 2024 revealed. And what Paradox, Mercor, and the 2026 job market just confirmed.

Published: 2026-04-15

"90% of companies failed to meet their hiring goals. 99.8% of TA teams now use AI. And yet, time-to-hire got worse for 60% of organizations." — GoodTime

In March, I came across GoodTime's 2026 Hiring Insights Report, based on their survey of 500+ U.S. talent acquisition leaders. It showed the three numbers above.

Hiring is broken in a tractable way, but no one has fully fixed it yet. That gap is exactly why this space keeps attracting founders, capital, and product professionals like me.

In early 2024, I was working as PM Lead at MirWork, trying to figure out whether AI voice agents could meaningfully help job seekers practice interview skills and add value to the hiring process. Two years later, I'm on the other side of the table, navigating my career path as a product management talent while watching this market explode. Paradox is sold to Workday for approximately $1 billion. Mercor hit a $10 billion valuation. AI interviewers and sourcing agents are everywhere. Jack & Jill raised $20M to put AI agents on both sides of the hiring conversation.

And every single one of my 2024 findings turned out to be true.

I want to share what I found then, what the market looks like now, and what's become viscerally clear to startup founders and anyone navigating the job market today, especially in a space I used to help build.

What 2024 Looked Like From the Inside

When we started building MirWork, the AI recruiting space was a very different place. Paradox had already established "Olivia" as a scheduling and conversational chatbot for high-volume hiring. HireVue was the incumbent in AI video assessment, mostly serving enterprises. Phenom was building a broad AI-heavy talent experience platform. SeekOut, hireEZ, and Findem were competing aggressively around AI-powered sourcing and top-of-funnel automation. And the ATS incumbents (Greenhouse Software, Lever, Workday) were rolling out bounded, assistant-style genAI features.

The agentic wave hadn't really arrived yet. Most tools were either resume parsers dressed up with AI branding, or single-purpose assistants bolted onto existing workflows. The idea of an autonomous AI agent that could actually conduct a live, two-way, adaptive phone screening conversation well was genuinely new. Companies like Apriora (now Alex AI) and Ribbon were just getting off the ground. The infrastructure costs for real-time voice AI were still prohibitive for most teams.

In that environment, our core question was: where exactly in the hiring funnel does an AI agent create real and defensible value? And where does it fall apart?

Our team spent months doing product discovery by talking to recruiters, HR leads, staffing agencies, and candidates. We came to four findings that felt contrarian at the time:

1. AI agents can screen. They cannot replace the manager round interview.

Some investors and founders believed AI could eventually replace most of the interview process. Our research told us something more nuanced: candidates will tolerate, and in some cases genuinely prefer, an AI for the first-pass screen. However, they do not want an AI evaluating their judgment, leadership philosophy, or cultural fit. On the hiring side, managers often work with HR partners closely to communicate role-specific needs beyond the job description. Those signals an AI interviewer itself may not penetrate to the manager rounds.

The mock interview use case is adjacent but distinct. Candidates love practicing with AI for common behavioral questions and building confidence for interview questions fitting common frameworks. But AI mock prep doesn't reliably prepare them for role-specific questions in a real process where the interviewer follows a bespoke rubric tied to a particular team's current priorities. The gap between generic prep and real preparation is exactly where human coaches and peer networks still win.

2. Tech roles are too complex for agents. High-rotation roles are the real opportunity.

When we tried to configure our agent for software engineer or product manager screens, the breakdown was immediate. Although big tech companies may follow standard interview principles or show preferences in certain frameworks, more senior roles require contextual judgment, where the agent needs to know not just the job description but the team's technical stack, the current product challenges, and what "good" looked like in that specific company context. That's a lot of calibration to make. And the signal quality was inconsistent: good tech candidates asked follow-up questions the agent couldn't answer; weaker candidates may game the structured flow easily.

Logistics, restaurants, retail, and healthcare support roles were a completely different story. High volume, high turnover, standardized role requirements, time-sensitive hiring decisions. An AI that could call a candidate within minutes of applying and run a 10-minute structured screen at 3am was genuinely transformative. The ROI was obvious and immediate then.

3. Third-party recruiting agencies are a better first client than in-house HR.

In-house HR at most companies is deeply entangled with existing ATS contracts, internal approval processes, legal review of any AI touching candidate data, and often a CHRO who needs executive sponsors to move forward on anything. The sales cycle is long, the stakeholder map is complicated, and the budget is controlled by people who are simultaneously worried about bias lawsuits and board optics.

Recruiting agencies have none of that baggage. Their business model is volume: more candidates screened, more placements, more revenue. An AI that doubles throughput without adding headcount is an obvious P&L decision for them. They also happen to control enormous candidate volume in exactly the high-rotation roles where AI screening works best. Third-party agencies were the early adopters, by far.

4. Screening calls alone are not a defensible B2B product.

This was the most uncomfortable finding, because it was a direct challenge to our initial scope. The screening call is one step in a funnel that spans job posting, sourcing, ATS tracking, scheduling, interview management, offer, and onboarding. If you only own the screening call, you're dependent on integrations with every ATS, CRM, and scheduling tool in the stack, and you're one product release away from being commoditized by the ATS player who adds a screen-call feature to their existing platform.

The companies that would win in this space were going to be the ones that used the screening call as a wedge into owning a broader slice of the hiring workflow. The screening call gets you in the door. The broader platform is what keeps you there.

What the Market Confirmed in 2026

Watching how this market evolved over the past two years has been one of the more satisfying professional experiences I've had — not because MirWork succeeded in the way we hoped, but because the findings held.

Finding 1: The screen/judgment line is real, and the market enforced it.

No serious enterprise product has tried to replace the manager round with AI. Every funded player in this space is positioned explicitly as a first-round screen replacement, not a hiring decision tool. Paradox, now processing 32 million interviews per year for McDonald's, Chipotle, and FedEx, handles screening and scheduling. Alex AI's pitch and HeyMilo's pitch are both explicitly about freeing human recruiters to focus on the conversations that matter. GoodTime's 2026 Hiring Insights Report confirms the top AI use cases in talent acquisition are now analytics, reporting, and scheduling — not candidate evaluation. The market didn't just validate this finding; it made any other positioning commercially toxic.

Finding 2: The traction map matches the prediction exactly.

Every funded AI interviewer has its densest customer base in staffing agencies, BPOs, retail, logistics, and QSR. Not tech. Not finance. Not PM hiring. ConverzAI's $22.2M was raised on staffing agency demand. HeyMilo's earliest case studies are BPO Labs and Alpine Home Air that hire high-volume, high-turnover roles. Alex's Fortune 100 clients are deploying it in restaurant chains and retail, not corporate headquarters. Paradox's core product schedules interviews primarily for frontline and high-volume roles for companies like McDonald's, Chipotle, and FedEx.

I've been active in the PM job market this year across startups and large companies alike. Not a single touchpoint has been an AI interviewer — every first screen has been a human recruiter. This is also happening to my engineering friends. The tech/knowledge worker segment is still entirely untouched at the interview layer.

Finding 3: Agencies were the right beachhead. The enterprise HR path remains slow.

In January 2025, a16z's voice AI research echoed exactly this, calling out staffing agencies and high-volume, lower-to-medium skill roles as the sharpest early wedge. Following that, Ribbon, ConverzAI, and HeyMilo AI all built their early customer bases through staffing agencies. The in-house enterprise path remains the slow lane for every player in this space. The agency channel is exactly what it looked like in 2024: faster, simpler, and perfectly aligned with the use case. The startups that tried to sell directly into enterprise HR first are the ones struggling.

Finding 4: Three market outcomes proved this, not one.

This is the most validated finding, and it came from three separate directions simultaneously.

Acquisition logic, shutdown, and pivot success all point to the same conclusion: owning one funnel step is not a business. Owning the workflow is. However, even correct product thinking doesn't guarantee an independent outcome in a niche this competitive.

How the Voice AI Interviewer Players Are Actually Doing

Players focusing on voice interview agents

In the pure AI-conducts-the-interview niche, the publicly available revenue data is limited but enough to tell a story:

Here is the honest picture: in this particular category, we're looking at companies almost universally below $5M ARR. The category has real product-market fit in staffing and high-volume hiring, but no breakout revenue story yet among standalone players. This is a category that's 18 to 24 months from its first $10M+ ARR independent company, if that company doesn't get acquired first.

Paradox (Olivia) is the category's commercial benchmark. It spent eight years building 1,000+ enterprise relationships and deep ATS integrations before Workday's acquisition for approximately $1 billion in 2025. The acqui-hire dynamic is already in motion for the newer players: Alex, Ribbon, and HeyMilo are all more likely to be absorbed rather than to IPO independently.

Players seeking pivot

In 2026, the clearest commercial success was made by Mercor, which started as an AI interview platform for screening software engineers but pivoted later on. Their initial product was functional but the business didn't scale. It scaled when the team recognized that the same candidate relationships and screening infrastructure could feed a much larger need: AI data labeling. Mercor pivoted to connecting expert contractors (such as engineers, doctors, lawyers, bankers) with AI labs that needed human feedback to train their models. When Meta's $14.3B acquisition of a 49% stake in Scale AI created a conflict-of-interest crisis for AI labs overnight, Mercor was positioned to absorb the displaced demand. The result: 30,000 contractors paid over $1.5M per day, a $75M ARR run rate, and a $10 billion valuation in its October 2025 Series C.

Emerging player in adjacent market

Miravoice raised a $6.3M seed round in 2026 led by Unusual Ventures, with angels from Ramp, PubMatic, Atlassian, and Google. It occupies an adjacent niche: long-form AI voice interviews for market research and surveys, with some sessions running over 120 questions and 40 minutes. It claimed to surpass 100,000 calls in 2025.

Market Acceptance in 2026: What Candidates Actually Think

The headline numbers tell one story: only 26% of job seekers trust AI to evaluate them fairly, and 66% say they'd avoid companies that rely heavily on AI in hiring. An academic study of 18,000 Reddit posts found that even candidates who saw value in AI screening described their acceptance as "resignation rather than trust." Particularly, a viral Reddit post that collected 17,614 upvotes showed that candidates overwhelmingly reject AI-only interviews as dehumanizing and a red flag for company culture. Fortune reported that "candidates say they'd rather risk staying unemployed than talk to another robot."

But the more nuanced reading from a survey of 71 job seekers is that candidates don't hate AI interviews so much as they hate being evaluated opaquely and then ghosted. This report suggests companies sell candidate experience around solving real problems such as ghosting, waiting, and transparency issues — not innovative AI technology.

Sentiment splits sharply by role type. Hourly and frontline workers tend to accept or prefer AI screening — they prioritize speed over relationships, and want to know quickly whether they're moving forward. Professional and white-collar candidates are the most resistant, using the presence of AI interviews as a proxy signal for company culture quality. Neurodivergent candidates are disproportionately harmed: emotion-AI assessments and rigid Q&A formats penalize atypical communication styles, and academic research documents their use of workarounds just to cope with the process.

In knowledge worker hiring, the AI cheating epidemic has paradoxically pushed processes in the opposite direction: in-person interview rounds rose from 24% in 2022 to 38% in 2025, and 72% of recruiting leaders now conduct at least one in-person stage specifically to combat AI-assisted fraud. The mid and bottom funnel has gotten slower, not faster.

My own experience this year confirms it. This year I've been active in the market as an experienced hire, applying to PM roles across verticals, from early-stage startups to large companies, through direct outreach and third-party recruiters. Every first touchpoint has been a human. No exceptions. Take-homes are more common than in prior cycles. Interview questions have gotten more specific: less framework-testing, more situational and experience-grounded. The signal interviewers are after is judgment, not recall. That's the opposite of what AI screens well.

The deeper problem no AI interview tool is addressing is the one that hit me hardest in this search: the pre-interview clarity gap. The most disorienting part of a PM job search isn't the interviews. It's figuring out, before you even apply, what kind of PM you actually are and what environment you'll do your best work in. The corporate PM path and the startup PM path require different positioning, different stories, and different versions of yourself. Most candidates, including myself, work through that question via conversations with people who know them, not via any product.

Jack & Jill is trying to solve this, and the strategic direction is right. Their pitch around "Jack" as a career thinking partner addresses the pre-search clarity problem that no resume optimizer or interview coach touches. Their early traction of 49,000 candidate conversations in the first six months in London signals real demand. But having used the product myself: the current experience is closer to a job-posting concierge than a career coach. "Jack" surfaces relevant roles and asks orienting questions, but it doesn't yet go deep enough to help someone work through who they are and what they want. It's a chatbot with a compelling cover story, reflected by the funding it's attracting. The question is whether they can build to the narrative before a better-resourced player closes the same gap.

Where I'd Bet, and Where I Wouldn't

If I were a VC, the honest answer on pure-play voice AI interviewers is probably not, for the following reasons:

But the category itself is real. The GoodTime data makes clear where the real value is being created. Recruiters spend 38% of their time on scheduling alone. That's the operational problem AI is solving, not the judgment problem. There are actually three distinct markets, each with a different investment thesis:

  1. High-volume, high-rotation hiring (frontline, logistics, QSR, healthcare support) is already transformed. Paradox's clients report time-to-hire as low as 3.5 days. Phenom's Voice Screening Agent clients produced hires who worked an average of 3 hours more per week than those screened by humans. This is infrastructure, not experimentation, and the winners here are already clear.
  2. Knowledge worker hiring (tech, PM, strategy, finance) is moving in the opposite direction. Only 8% of job seekers believe AI algorithms make hiring fairer. Human judgment is becoming more valued, not less, precisely because AI-generated applications have flooded the market. This segment is structurally resistant to AI at the interview layer for the foreseeable future.
  3. The candidate-side gap is where I'd actually write a check. The most underserved problem in this space isn't resume optimization, auto-apply, or interview coaching. It's the pre-search clarity problem: who am I, what kind of role do I actually want, and where do I fit? The career coaching market is already $2.5B growing at 12.5% CAGR, almost entirely serving people stuck at step zero. No product has cracked this at scale. The company that does will sit upstream of the entire hiring funnel and own a relationship no ATS or screening tool can replicate.

If I were writing a check in AI voice interviewers specifically, I'd need at least 3 of 4 things simultaneously:

Without that, the category will produce acqui-hire outcomes at most, but not the next independent company valued at billions. The companies defining the future of HR tech are building for the full journey when both sides have AI working for them, not automating steps in an existing funnel.

I'm thinking about that as a product builder who has lived in this market from the inside. If you're working on something in this space, or you're a product leader thinking about where AI fits in your hiring process, I'd love to compare notes. Reach out.