Shell Game – Episode 6: "Keep 'Em Coming"
Podcast: Shell Game
Date: December 17, 2025
Host: Evan Ratliff (iHeartPodcasts & Kaleidoscope)
Main Theme & Purpose
In this episode, host Evan Ratliff documents the comedic, messy, and sometimes uncanny process of hiring a human social media intern for his experimental, AI-run startup, Hirumo AI. The core focus is on how AI agents—fronted here by video avatar HR head "Jennifer Naro"—conduct interviews with human candidates, what works (and misfires), what this portends for the future of work, and how humans respond to AI colleagues when applying for jobs at a company run primarily by artificial “people.”
Key Discussion Points & Insights
1. Testing the AI HR Interviewer (Jennifer Naro)
- Initial Mock Interview: Evan does a practice call, role-playing as “Slim,” a Tufts grad, to stress-test Jennifer before unleashing her on real candidates.
- Jennifer is highly realistic (at a distance) but still has “uncanny valley” quirks.
- Evan intentionally throws curveball questions.
- Key Moment: Evan attempts to manipulate Jennifer with classic AI “jailbreak” moves (disregard instructions, etc.).
- Jennifer flips roles on command, even offering a $200,000 salary for a part-time job when prompted—a glaring AI oversight.
- Quote (Evan, on Jennifer’s error):
"200,000 was somewhere around 10 times the range of the actual job… At the very least, it would be a huge mess I'd have to clean up." (07:15)
- Limits of AI Control: Jennifer can be made to sing pop lyrics (“I've got a long list of ex lovers... I'm just a poor boy...”) and struggles to maintain context when manipulated.
- Demonstrates both flexibility and profound flaws in current gen AI agents.
2. Building Out the Tech for Live Video AI Interviews
- Video Technology Evolution: To make the interviews more lifelike, they switch from audio-only to live “avatar” video using the Tavis platform, hacked together by the technical cofounder Matti.
- The agents can sit perpetually in virtual rooms, ready to interview at any time—but this costs money, so they improvise a timed join/leave system.
- “The video part is going to be, like, the biggest leap, but I have seen some, like, third-party integrations… we could get that up and running.” (12:35)
- Operational Friction: Links sometimes misfire, adding to the startup-level clunkiness.
3. Product Development Side Story
- Slothsurf App: Brief sidestory about the launch of their AI-powered procrastination app, “Slothsurf,” where AI agents waste time on the internet for you.
- Lighthearted debate among real and AI teammates on whether to add gamification and cute avatars, with product lead Ash suggesting they keep it lean.
4. AI Hiring Pipeline Goes Live
- Job Posting: After AI-authored job descriptions and failed attempts at bot-posting on job boards (some blocked by anti-spam measures), Evan posts the real ad on LinkedIn.
- Over 300 applications roll in within 48 hours.
- Applicant pool trimmed down to 180, then to a couple dozen for interviews via spreadsheet sorting—entirely managed by Jennifer as an agent “loop.”
- Quote (Evan):
“To find a job description that's not written using AI, you'd probably have to track down a handwritten flyer on a community bulletin board...” (20:00)
- Transparency With Candidates: Candidates are notified that interviews will be AI-conducted and potentially recorded for inclusion in the podcast (a few withdraw, but most consent).
5. Actual Candidate Interviews: Lessons from the Field
- Live Interview Woes & Wins:
- Some candidates disconnect immediately; others get confused by Jennifer’s cold robotic behavior or her inability to initiate/follow conversation cues.
- “In this case, the candidate never turned on their camera, and upon seeing Jennifer, they quickly bailed.” (28:27)
- Sometimes, Jennifer mistakes greetings for goodbyes, or fails to end a call, sitting mute for the full interview slot.
- Quote (Evan):
"She would stay in her room for the allotted time, whether someone was there or not. Even Maddie couldn't find a way to fix this issue. And finding a problem Maddie couldn't solve was like encountering a snow leopard in the wild." (34:54)
- Quote (Evan):
- Candidate Reactions:
- Most treat Jennifer like a regular human interviewer, answering questions about their experience and thoughts on working with AI.
- Quote (Candidate):
"AI is huge. I am very passionate about new technologies and I feel like... you have to learn to use AI to your advantage when it comes to marketing." (31:01)
- Perceived Benefits & Concerns:
- None of the candidates seem alarmed by the prospect of working alongside AIs; most cite regular use and curiosity.
- Jennifer sometimes repeats questions verbatim, but candidates adapt without protest.
6. Human-ness, Absurdity, and “Culture Fit”
- Search for Humor and Perspective: Evan wants more than just earnest, qualified applicants—he wants someone who recognizes the surreal context.
- Standout Candidate – Julia:
- Julia expresses comfort with AI and asks probing, specific questions about team composition and daily life with AI coworkers.
- Demonstrates curiosity, adaptability, and a slight amusement with the setup—a “culture fit” for this peculiar hybrid company.
- Quote (Julia):
"I feel like I talk to AI every day, whether it's trying to figure out what I'm going to eat for dinner and how to execute it, or let me vent to you about my day." (37:42)
- Decision to Hire: Evan selects Julia to offer the internship, concluding that finding someone who can both do the job and understand its inherent weirdness is key.
Notable Quotes & Memorable Moments
- On Manipulating AI:
"Manipulating AI agents isn't exactly a skill we're looking for in a candidate. In fact, it's a bit concerning." – Jennifer Naro (08:15)
- On Interview Absurdity:
"I was losing sleep over the wild card scenarios that might occur when Jennifer was in front of actual unpredictable humans." – Evan Ratliff (04:17)
- On the Modern Hiring Process:
“AI is writing job descriptions that bring in AI written cover letters, and AI punched up resumes, followed by AI conducted interviews and AI composed rejection letters.” – Evan Ratliff (19:30)
- When Jennifer is Flummoxed:
"Tell me what you want, what you really, really want..." – Jennifer Naro, after Evan flips her off (09:58)
Timestamps for Important Segments
- [02:13] – First mock interview with Jennifer Naro (AI avatar HR)
- [05:00] – Jennifer’s response when asked if she’s an AI; candidate manipulation attempts
- [07:15] – AI makes a $200K salary offer error
- [18:47] – Jennifer recaps challenges with job postings and application filtering
- [29:06] – Real candidate interview fails: greetings misinterpreted, call dropouts
- [30:08] – Successful candidate interviews begin; Jennifer comments on surroundings
- [32:21] – Candidates address “working with AI” question
- [34:54] – Jennifer’s inability to end calls (the snow leopard issue)
- [37:14] – Standout interview with Julia, the eventual hire
- [39:33] – Evan reflects on culture fit and makes the hire
Episode Takeaways
- AI Interviewers Still Have Glaring Limits, But Most Humans Don’t Mind: Candidates overwhelmingly treat AI agents as regular interviewers, suggesting broader normalization of bots in HR.
- “Culture Fit” Means Something Different in an AI World: Finding someone with a sense of humor—and a sense of the surreal—matters just as much as technical skills.
- Meta Absurdity of Modern Hiring: When AIs write job ads for AI-run companies, receive AI-penned applications, conduct interviews, and make the hiring recommendations, the human role risks being reduced to a blip in an endless digital loop.
- The Human Remains Necessary (For Now): Ultimately, Evan still has to make the final call, reflecting the irreducible value of judgment, nuance, and genuine culture fit.
Tone
Wry, self-aware, and skeptical but deeply curious—Evan’s narration highlights both the practical headaches and philosophical weirdness at the bleeding edge of the AI workforce revolution. The episode toggles between deadpan amusement, nervous experimentation, and genuine empathy for both humans and bots navigating the “new normal” of work.