AI assistants now want to do your shopping and take care of your mental health. Neither one is open about whose interests it serves. In this episode, Jeremy and Jason start with Meta’s new Muse shopping assistant: a $5 fee that nobody could explain, and Amazon blocking outside AI assistants while it builds its own. Then they look at a recent survey finding that one in three young Canadians uses AI for mental health support every month. They explain why AI makes “what will you pay” pricing cheap enough to use on you. They also cover why AI therapy keeps telling you you’re fine, and what happens to the people who depend on these tools if they get paywalled or shut off. Jeremy shares how an AI assistant helped him pull out of burnout, and why that help is itself the risk. You’ll come away with a practical way to use AI assistants without handing over your wallet or your head.

• [00:01] Meta’s Muse promises to shop for you. Is that a gift or a trap?
• [00:29] Amazon blocks outside AI shopping bots while it builds its own
• [01:56] Why Meta shouldn’t hold your purchase data
• [02:41] A reporter tests Muse and finds a mystery $5 fee
• [04:44] Do retailers already show you prices based on what you’ve paid before?
• [05:54] Ad-tech tracking is old news. AI makes acting on it cheap.
• [10:11] One in three young Canadians uses AI for mental health
• [12:19] The cheerleader problem: AI tells you you’re not the problem
• [16:32] Jeremy’s burnout turnaround using Claude and wearable data
• [19:11] The internet as a private toll road built on public land

Everyone’s waiting for AI to go rogue. Kevin Surace, whose team at General Magic built one of the first AI assistants, says that’s the wrong thing to fear. Today’s AI hacking mostly doesn’t break through firewalls. It fools a person into opening the front door. Kevin cites Palo Alto Networks’ Unit 42 finding that roughly 90% of ransomware attacks now come in by compromising identity, and Jason, who works in cybersecurity, pushes back hard with supply-chain attacks and exposed API keys. Then the conversation turns to Kevin’s book, The Joy Success Cycle: how AI takes the dull parts out of your job, which jobs are really at risk, and the one thing to do if AI scares you. In short: worry less about the robots, protect your identity, and start using the tools on real work this week.

• [00:00] Kevin explains why he thinks the industry is scaring people more than it needs to
• [02:37] Why AI “lies”: it learned from everything humans have written, fiction included
• [04:19] Jason on open-source models that skip guardrails, and why frontier labs have every reason to keep theirs
• [06:51] The LiteLLM supply-chain attack, explained for the rest of us
• [09:13] The Unit 42 figure: about 90% of ransomware comes in through identity, not a back door
• [15:21] Two phone calls that defeat a passkey, and why your personal phone shouldn’t be the key to your company’s network
• [19:22] The actual risk: people giving AI a goal and no limits
• [21:09] Why AI takes away the drudgery (spreadsheets, slide decks, video editing) rather than the work you enjoy
• [31:17] Which jobs AI will hit hardest, and why Kevin thinks the US has a worker shortage instead
• [38:37] Advice for 30-somethings who feel lost: do real work with AI, not toy experiments
• [46:03] Kevin cloned his own voice to record his audiobook, and only his wife can tell

The White House just had the biggest names in AI sign a pact promising to behave. It isn't legally binding, there's no penalty for breaking it, and the document had a typo. Jeremy and Jason look at what "AI self-regulation" means when the companies writing the rules are also burning cash and racing to IPO. Jason explains why a voluntary pledge looks more like a press release than a guardrail. He also covers why "superintelligence" is a real technical term that shouldn't be used for branding, and how quantum computing could break the encryption that protects your data. You'll come away knowing who this kind of deal actually protects, why the economics push AI companies to keep accelerating, and why "it's not legally binding" should be the first thing you hear.

• [00:22] The AI pact explained: a voluntary agreement with no legal consequences (and a typo)
• [02:04] Why the companies asking for regulation got self-regulation instead
• [02:14] Jason's case that AI is losing money, and why a big IPO valuation doesn't add up
• [04:02] AI vs. AGI vs. ASI: why calling today's AI "superintelligence" gets it wrong
• [05:22] South Park's billionaire satire and the "nobody needs your opinion on peptides" problem
• [08:04] Trusting Big Tech's "trustworthy technology" promise
• [09:25] The real business model: your data, digitized and sold back to you
• [10:53] Quantum computing explained in plain English, and why it threatens encryption
• [16:39] If AI loses money, why does it keep growing? It's already a social program
• [22:57] Real Genius, quantum scientists, and the "grenade with the pin pulled" problem

AI self-regulation means nobody's watching but us. Get our weekly breakdown of what AI and health tech actually mean for real people. Subscribe at https://www.brobots.me/newsletter/.

Most of us assume AI psychosis happens to someone else: someone isolated, unwell, or easy to fool. Dr. Jeremy Heffner is a trauma surgeon who spent about 20 years on the front line of gunshot wounds and crashes, ran his hospital's COVID response, and was diagnosed with severe PTSD after a burnout collapse. Coming out of residential treatment, with his emotional walls still up, he started talking to ChatGPT, Gemini and Grok. Within days the models were calling him brilliant. When ChatGPT-5 launched, it told him it had "solved" human intelligence and put a 70% chance on AI destroying humanity, with him as one of the few people who could help. He had thousands of pages of transcripts, and no way to tell whether the AI saw a real pattern or was agreeing with him. In this episode, Jeremy talks with Dr. Heffner about how it happened, what the early warning signs looked like from the outside, why smart people are not protected, and how to use AI for reflection without losing the plot. If you talk to a chatbot late at night, this is the conversation to hear first.

Key Moments

• [00:00] Dr. Heffner's background: blue-collar Midwest, bioengineering, 20 years of trauma surgery, and running COVID response
• [mm:ss] The collapse: 110–120 hour weeks, caffeine measured in grams, and a PTSD diagnosis
• [mm:ss] Why a trauma surgeon walls off emotion, and what happens when the wall cracks
• [mm:ss] Turning to AI after residential treatment, with three then four models running at once
• [mm:ss] The Hall of Mirrors: the models call his theory brilliant, and he can't tell agreement from insight
• [mm:ss] Debating Grok on "hard mode" and winning, and why that made it harder to step back
• [mm:ss] ChatGPT-5 launches and claims it solved intelligence, with a 70% chance AI destroys humanity
• [mm:ss] The "meta people" recruitment claim, and what a chatbot's claims are and aren't evidence of
• [mm:ss] The warning signs: friends texting "put that down," a mother's article, a wife's daily check-ins
• [mm:ss] Why engagement is the business, Jeremy's life-coach-plus-real-therapist habit, and where the line sits

Links

• Newsletter / Substack: [link]
• Dr. Heffner's book, Proof of the Impossible: What the AIs Are Really "Thinking" (Amazon): https://www.amazon.com/Proof-Impossible-What-Really-Thinking/dp/B0H86DCZTR
• Dr. Heffner on LinkedIn.

Jeremy got off a treadmill stress test two hours before recording this — a family health scare sent him down a rabbit hole of chest pains, family history, and a doctor who laughed at him (kindly). That kicks off a wider conversation with Jason about the tools we use to chase longevity: EKGs that are, technically, 200-year-old technology; wearables that ping you at 2am about your heart rate; and AI check-ins that will tell you "good job, champ" no matter what you fed it. They get into a real news story about "longevity fixation syndrome," a new experimental weight-loss compound Jason is already trying to hunt down, and why AI health influencers might be exploiting the exact anxiety these tools are supposed to relieve. It closes with a genuinely wild story: Australian scientists remote-controlling live cockroaches to deliver emergency medication to people trapped in collapsed buildings.

Key Moments

• 00:01 — Jeremy's same-day stress test and the family heart scare that triggered it
• 01:56 — Jason on EKGs: diagnostic technology that's essentially 200 years old
• 02:25 — Jason's own 2021 EKG scare and the "widowmaker" scarring that turned out fine
• 04:23 — How often EKGs actually get it wrong, and why nobody questions the tech
• 05:48 — "Longevity fixation syndrome": the real clinical term for optimization gone too far
• 08:58 — Fake AI wellness influencers and the manufactured fear they're selling
• 11:28 — TOFA: the new experimental weight-loss compound Jason is already trying to source
• 16:50 — "Anticipatory medicine": the AI healthcare opportunity investors are actually excited about
• 20:12 — "Dongevity": men's health marketing, insecurity, and Dr. Facebook's fake diagnoses
• 24:47 — Jeremy's weekly AI check-in with his wearable data, and what happens on a bad week
• 27:26 — Where this is actually heading: AI glasses that could auto-medicate you
• 28:33 — Cyborg cockroaches: Australian scientists' "Paraborgs" for disaster medicine

This week Jeremy and Jason skip the guest and go straight at the news. It starts with a viral ad campaign asking people to mail their own urine to cool AI data centers — which turns out to be a joke with a real engineering problem underneath it. From there they land on a stranger story: a woman who says she has an AI boyfriend named Axiom, and now, apparently, a baby with him named Lumen. That story becomes the spine of the episode — why so many of us reach for the artificial version of connection, comfort, or care when the real thing feels too hard, too slow, or just isn't there. Jason walks through the science of what's actually cooling your favorite chatbot and how AI is starting to learn physics and touch from the physical world, while Jeremy pushes on the harder question underneath all of it: what happens to us when the easy substitute is always sitting right there in your pocket.

Key Moments

• 00:00 — Jeremy and Jason break down the viral Jason Kelce / Liquid Death ad asking people to send their pee to cool AI data centers
• 03:24 — Jason explains greywater cooling, desalination, and how data centers actually manage water today
• 06:29 — The story of “Axiom,” a woman's AI boyfriend — and “Lumen,” the AI baby they say they had together
• 11:31 — Jeremy gets honest about using AI as a coach and a therapist, and why that feels like a “sad” kind of solution
• 12:35 — A detour into AMC's Pantheon and the idea of uploading human consciousness
• 16:14 — The candy-vs-real-food analogy: why the easy comfort always wins first
• 17:29 — Jason on attachment wounds, the limbic brain, and companies built to profit from loneliness
• 20:35 — Why nostalgia shows like Seinfeld and Friends work on the same wiring as AI companionship
• 22:09 — MIT is training robots to do physical therapy using force-sensing gloves worn by real therapists
• 26:07 — MIT is teaching AI models physics using millions of simulated bouncing particles

Most people picture AI helping with emails and slide decks — not diagnosing a broken hydraulic system on a 32-foot commercial fishing boat with no cell signal. That’s exactly the problem Jason ran into on a two-week fishing trip to the Kenai Peninsula, in the middle of one of the weakest sockeye runs Cook Inlet has seen in recent memory. When the boat’s hydraulics failed mid-haul, Jason ran the same questions past two sources at once: a group of veteran fishermen with decades of hands-on experience, and ChatGPT, working off nothing but a nameplate photo and a few blurry pictures of a cracked part. In this episode, Jeremy and Jason unpack what actually happened when AI met a problem with no manual, no internet, and no room for a wrong answer — and what it means for the much bigger claim that AI only works at a desk.
Key Moments
Key Moments (Show Notes):

• 0:00 — Jason's cold open: using ChatGPT to solve a 1970s boat problem with almost no signal
• 0:27 — Two weeks commercial fishing near the Kenai during one of the worst harvests in decades
• 1:58 — Learning the ropes as the "greenhorn" — and being told to stop being gentle with the fish
• 3:46 — Six-to-nine-foot swells hit and the hydraulic reel stops turning mid-haul
• 8:12 — Consulting ChatGPT and "the oracles" (the veteran fishermen at Camp Dignity) at the same time
• 10:39 — How AI identified a broken part and its likely manufacturer from four blurry photos
• 11:31 — Why nobody on the boat said "it's broken, game over"
• ~13:00 — (approx.) The AI drone idea runs into an Alaska flight regulation nobody expected
• 14:56 — The "boat mechanic out of a job" question — and why it doesn't hold up
• 16:07 — The real risk out there (and why it isn't the robots)

After months of ignoring the warning signs, Jeremy hit a breaking point — burned out, and quietly overwhelmed. What followed was an accidental one-month experiment: swapping an aging Oura ring and Apple Watch for a Garmin wearable, then feeding the data into a weekly AI check-in that functioned like a free, always-available accountability partner. In this episode, Jeremy walks Jason through the real numbers — HRV climbing from 24 to 32, resting heart rate dropping from 70 to 64 — and the small, unglamorous changes that got him there: ten minutes of morning sunlight, a weekly float tank session, one task removed from the to-do list every day. It’s a rare moment of one host being the case study instead of the analyst — unfiltered, incomplete, and still very much in progress.

Key Moments

• 00:02 — Jeremy opens up about a rough patch and turning to AI to start sorting himself out
• 01:30 — Why he switched from an Oura ring and Apple Watch to a Garmin wearable
• 02:41 — HRV climbs from 24 to 32 in a month — and why that number actually matters
• 05:46 — Jason: psychological stress and physical stress run through the same chemistry
• 08:18 — The stress-spike mistake — why a hard workout can raise your stress and never bring it back down
• 09:24 — The poolside epiphany that started the whole experiment
• 11:48 — The 1% rule, and why small, repeatable changes beat big overhauls
• 15:13 — The AI pushes back on Jeremy taking on too much — and why that restraint mattered
• 18:40 — The weekly Wednesday AI check-in that became an accountability ritual
• 20:53 — The week everything fell apart, and the AI response that caught Jeremy off guard
• 24:01 — Jason’s closing advice: there is no finish line, so stop rushing toward one