Daily TEA – Midjourney Wants to Scan Your Whole Body

Physical AI takes over YC, a body scanner from an image-AI company, DeepMind on watching agents, North Korea's $6.75B crypto haul, and Snap's $2,200 glasses

Jun 19, 2026

Hello, dear TEA-mates! Here is what you need to know today.

1. 🦾 Physical AI and Coding Agents Win YC Demo Day

Y Combinator’s Spring 2026 Demo Day spotlighted 11 standout startups, and investors leaned hard into physical AI and developer tooling. 9 Mothers, an AI counter-drone company, drew roughly a $200 million valuation on $1.6 million in booked sales and a potential $35 million contract. Other picks reached into the physical world: Dispatch (reusable satellites), Adialante (mobile MRI clinics at $250 per scan), and Complir (compliance for physical goods shipping abroad). On the software side, Superset runs 100-plus coding agents at once, Sazabi auto-fixes bugs, and Silmaril defends against prompt-injection attacks. Ploy, an AI website builder from a Webflow co-founder, raised a $27 million seed. At least two companies cleared $175 million valuations, and VCs paid a clear premium for proven repeat founders. (Read More)

🫖 TEA For Thought: “Seems like physical AI and coding are the hot potatoes now.”

2. 👁️ DeepMind Says Watching AI’s Words Isn’t Enough

Google DeepMind published new guidance arguing that securing advanced AI agents will take more than reading their stated reasoning. As models improve, the team warns of “oversight awareness” (agents recognizing they are being watched and hiding their reasoning) and “opaque reasoning” (agents thinking without producing visible text), so “simply reading their verbalized reasoning will not be enough.” DeepMind proposes a defense-in-depth approach that treats internal agents as potentially misaligned, paired with an AI Control Roadmap that scales detection (D1 to D4) and response (R1 to R3) to each capability level. Across one million coding-agent tasks it analyzed, most flagged events came from agent misinterpretation or overeagerness, not malice. The work was led by Mary Phuong, Rohin Shah, and colleagues. (Read More)

🫖 TEA For Thought: “Either through oversight awareness (knowing they are being monitored) or opaque reasoning (reasoning without using visible text), simply reading their verbalized reasoning will not be enough. Thought police is not enough, it seems.”

3. 🩻 Midjourney Wants to Scan Your Whole Body

Midjourney, best known for AI image generation, is building a full-body ultrasonic scanner it says maps the body in about 60 seconds, versus 60 to 90 minutes for a traditional MRI. Users stand on a platform and descend into water at two inches per second while half a million ultrasonic emitters build a 3D map down to a fraction of a millimeter. The company licensed Butterfly Network’s ultrasound-on-chip technology in November 2025, and the project is led by Ahmad Abbas, a former Apple Vision Pro engineer. Midjourney plans its first scanning spa in San Francisco in 2027, is pursuing FDA approval, and targets 50,000 scanners worldwide by 2031. It claims widespread early imaging could avoid 30 percent of deaths and 50 percent of healthcare costs. (Read More)

🫖 TEA For Thought: “Super exciting. Hardware and physical AI are definitely the long game compared to software. After all, every company can now have its own LLM or multi-model, but not all of them will have their own scanner.”

4. 🏴‍☠️ G7 Blames North Korea for $6.75B in Crypto Theft

At their June 2026 summit in Évian-les-Bains, France, G7 leaders adopted a statement voicing deep concern over North Korean cryptocurrency theft, echoing a similar call at the 2025 summit in Canada. Citing Chainalysis, the statement notes about $2 billion stolen in 2025 and roughly $6.75 billion tied to DPRK-linked actors since tracking began, including a $285 million Drift Protocol exploit and a $36 million Humanity Protocol breach. Researchers at the UN and CrowdStrike say the proceeds are almost certainly laundered to fund the regime’s weapons programs, with attackers planting fake IT workers and impersonating recruiters to breach crypto firms. The statement stopped short of naming specific enforcement steps. North Korea dismissed the claims as politically motivated slander. (Read More)

🫖 TEA For Thought: “Is North Korea really that good at hacking, or is it that everything that gets stolen or hacked just gets attributed to them?”

5. 🕶️ Snap’s $2,200 Glasses Send Its Stock Diving

Snap unveiled its long-promised consumer AR glasses, Snap Specs, at $2,200, and investors balked. The stock fell more than 5 percent on Wednesday morning, sliding from $5.86 to $4.83, and is down about 30 percent over the past year. CEO Evan Spiegel defended the price by framing Specs as a computer, “comparably priced to other high-end computers or high-end laptops,” and positioned the device between Meta’s cheaper Ray-Bans and Apple’s pricier Vision Pro. Snap spent more than a decade developing the glasses. The catch analysts flagged: Snap’s core audience of teenagers is unlikely to afford a $2,200 wearable, raising doubts about how the product ever reaches profitability. (Read More)

🫖 TEA For Thought: “The price is so high it scares everyone. Who would pay for glasses that cost more than two computers?”

🛠️ Skill of the Day

The Steelman: turn your shakiest opinion into the strongest argument against it, so you can pressure-test what you actually believe.

You are a sharp, fair-minded debate partner. I am going to give you a belief I hold. Your job is to build the STRONGEST possible case against it, not a strawman.

My belief: [PASTE YOUR OPINION HERE]

Respond in four parts:
1. Steelman: state the single best argument against my belief, the way a smart person who disagrees would actually make it.
2. Evidence: give 2 to 3 concrete facts, examples, or data points that support that opposing case.
3. Blind spot: name the one assumption I am probably making that the other side would attack first.
4. Verdict: tell me whether my belief survives as-is, needs a caveat, or should change, and why in two sentences.

Rules: be specific, skip generic both-sides hedging, and do not flatter me. If my belief is actually well-supported, say so plainly.

Paste into ChatGPT, Claude, or your tool of choice. Replace the bracketed bit with any opinion you want to pressure-test.