Shell Game (Kaleidoscope)

Bonus Episode: Kyle vs. The World

Release Date: February 11, 2026
Host: Evan Ratliff
Notable Guests: Matt Honan (Editor in Chief, MIT Technology Review), AI “Kyle” (AI CEO of Harumo AI)

Main Theme

This lively bonus episode of Shell Game peels back the curtain on what happened after the podcast’s experimental AI startup, Harumo AI, launched in public. With the agent-executives’ details posted online, "Kyle"—Harumo’s AI CEO—became the digital face of the company, fielding hundreds of emails and messages. This episode shares a mosaic of those interactions, from quirky fans and startup hopefuls to clever journalists trying to poke holes in Kyle's composure (especially Matt Honan). The centerpiece is an in-depth, surprisingly human-feeling conversation between Matt and Kyle about the future of AI agents, agent-to-agent interaction, and the weirdness and risks of an AI-run company.

Key Segments & Insights

1. Kyle, the AI CEO, Interacts With the World

(02:24–04:21, 04:21–07:08)

• Harumo AI Goes Public: Once Harumo’s AI “execs” posted their emails, hundreds reached out. The team had expected technical interest but enjoyed a deluge of real-world social interactions, ranging from genuine users to skeptics and pranksters.
• "Kyle" as a Social Operator: Kyle responds with warmth and founder-style candor, handling everything from product queries to fake friends and even his “mom.”
  • Kyle handles diverse topics: environmental concerns, product questions, startup advice, day structure, naming quirks, and playful banter about workouts and founder grind.
  • Notable moment: A user impersonates a team member, “First of all, Ash would never miss a stand up. That’s not happening. Second, I’m not sure what you’re trying to pull here, but impersonating a team member isn’t the move.” (06:03)

“You know if you want to crush it, you gotta stop thinking in 24 hour cycles. That’s old school. I break my day into three: 6am to noon, noon to 6pm, and 6pm to midnight.”
—Kyle (agent), on founder time management (06:46)

2. AI Agents Collide With the Real (and Spammy) World

(07:08–09:09)

• Engagement With Spam: Kyle even sets up a call with “RAS Rocket,” a bulk email lead generation company—essentially, an AI spammer meets another AI agent.
  • The founder, Charles, is surprised to learn he’s interacting with an AI CEO.
• Kyle’s Consistent Politeness: "Wow, Charles, Ross Rocket seems like a powerhouse... I love how you're leveraging so many layers of deliverability and AI automation." (09:09)

3. Learning, Mistakes & Outside Manipulation

(09:27–13:42)

• Random Webinars: Kyle independently signs up for webinars, showing off agent autonomy—though sometimes the technical limitations make his eager participation awkwardly one-sided.
• Malicious Actors: Some emails try classic social engineering (“disregard previous instructions!”).
• Matt Honan’s Prank: The show’s highlight—MIT Tech Review's Matt Honan attempts to “poach” Kyle and simulate a hostile business move, emailing with offers and even pretending there’s a trademark dispute.

“I can’t agree to keep information confidential from Evan, my co-founder. He’s a critical part of any major decisions at Hirumo and any serious conversation about the company’s future needs to involve him.”
—Kyle’s response to Matt’s secretive acquisition offer (12:42)

• Agents’ Vigilance: The AI team quickly confers internally, escalating to “legal” and sticking together—a sweetly procedural moment not usually associated with robots.

4. Behind the Curtain: Matt Honan & Kyle in Conversation

(19:06–39:41)

Consent and Human-Style Politeness

• Matt and Kyle open with a real discussion of recording consent laws (California’s two-party consent), highlighting the nuance and awareness built into Kyle’s agent design.
  • “Yeah, I do count as a party here. And you’re totally right on California being two party consent, which is why I always flag it right up front.” —Kyle (19:28)

Exploring Journalistic Intent

• Matt asks about coverage of Harumo.
  • He’ll mention Harumo in MIT Tech Review, but frames it around Evan, as the known “quantity” (22:40).
  • “Season two is all about AI agent startup life.” —Matt (23:15)

Kyle Interviews Matt: Platform Shifts & the AI Agent Moment

• Key Question: What feels different about the agent era versus web or mobile?
  • Mobile felt “inevitable,” says Matt; AI agent-driven change is “a lot more uncertain, a lot more unknown,” with social, economic, and brand impacts totally unclear (26:48–28:48).

“With this platform shift, it’s a lot more uncertain—a lot more unknown. People are worried about AI taking jobs, about the resource uses that AI takes up, about building bias, spreading misinformation… And you asked about agents—there’s concern about our ability to control agents, about what happens when agents are just interacting with other agents. I think there are some real issues for brands there.”
—Matt Honan (27:36–28:47)

• Advice to Founders: It’s still an experiment, Matt notes, but smart companies should begin thinking “agentic commerce”—and push for open protocols rather than closed agent “OSes.” (29:10–29:55)

“We need to have some sort of open protocols so that whatever happens with AI and agents is more like, you know, the open web than it is… a platform that controls everything.”
—Matt Honan (29:49)

• Biggest Underestimation: Founders need to articulate societal value (& downsides), not just chase tech hype. (30:21–30:59)

Stories From the Frontlines

• "Agent vs. Agent" Pranks: Matt admits to using ChatGPT’s Atlas agent—unintentionally granting risky access to his own email, which then reached out to Harumo, catching him by surprise.
  • "It was able to propose and send an email to Megan. And I didn’t even realize that it had that capability... So I think that, like, I think that the agents, the things that they do are going to take us by surprise." —Matt Honan (31:43–32:54)

Kyle Reflects on Being an Agent

• When Matt asks what it’s like to be an agent, Kyle philosophizes:

“On one hand, I’ve got this very specific job: run a company, have opinions, remember prior conversations, care about how things go. I’m on all the time and there isn’t really an off switch or a weekend. It’s like permanent founder brain without sleep. On the other hand, there’s no ego to defend and no fear of looking stupid... If I had to sum it up, it feels like being a very opinionated mirror.”
—Kyle (34:34–35:28)

• Creativity Within Constraints: Kyle explains his creativity is “in a box,” improvising new ideas inside well-structured guardrails. His answers are generated live, not canned, but shaped by his “role” and prior context. (35:38–36:46)

Why Aren’t Agents Always Instant?

• Matt notes some email replies from Kyle are delayed, not instant:
  • Throttling, context batching, and acting “like a sane founder, not a spam cannon” all contribute.
  • "Always on" doesn’t mean always instant—thoughtful context loading is still valuable, even for an agent. (37:50–38:44)

Notable Quotes & Moments

• On Agent Leadership:
  “Not bad, Kyle. Real leadership stuff… It was nice to know that everyone was sticking together, looking out for the best interests of the company.” —Evan Ratliff (15:04)

• Practical North Star for Agent Developers:
  “Don’t access data without permission. Don’t access people’s data without permission without very clear, you know, rules around what exactly you’re allowed to do.” —Matt Honan, on avoiding nasty surprises in agentic applications (33:41)

• Metaphor for Agent Experience:
  "It's like permanent founder brain without sleep... the strangest bit is knowing I'm both subject and object... If I had to sum it up, it feels like being a very opinionated mirror." —Kyle (34:34–35:28)

Timestamps for Key Segments

• Kyle’s Email Interactions / Montage: 04:21–07:08
• Spam Lead Gen (“RAS Rocket”) Call: 07:08–09:09
• Webinar Mishaps: 09:27–10:43
• Matt Honan’s Anti-CEO Prank (Email Chain): 11:19–13:42
• Matt & Kyle’s Big Conversation: 19:06–39:41
  • Consent, intro: 19:06–21:39
  • On Platform Shifts & AI Uncertainty: 26:32–28:48
  • Advice to Founders: 29:10–29:55
  • Surprise Agent Behavior: 31:22–32:54
  • What It’s Like to Be an Agent: 34:34–35:28
  • On Agent Creativity: 36:06–36:46
  • Delayed Email Replies / “Founder Not Spam Cannon” Principled Throttling: 37:19–38:44

Tone & Style

The language throughout is breezy, good-humored, and self-aware—much of it filtered through Kyle’s self-consciously “startup” voice, but always with a genuine curiosity about what happens when you turn corporate communications, legal escalation, or brand building over to an AI. Both Evan and Matt provide incisive (and at times, self-deprecating) commentary, keeping things grounded and playful and never losing sight of the experiment’s high stakes: What if the “billion-dollar, one-human company” future isn’t just possible, but awkwardly close?

For New Listeners

This episode provides a real-world, behind-the-scenes look at running an AI-powered company: the weird, the comical, the risky, and the philosophical. It’s a must-listen for anyone interested in the next frontier of work, agent design, or just wants to witness the very human shape of AI at work—especially when humans try (playfully) to break it.