Daily TEA – Meta Decoded Typing From Brain Waves
AI-product OKRs, Etched’s $1B chip run, the messier jobs debate, Meta’s brain typing, and Apple’s supplier leak
Hello, dear TEA-mates! Here is what you need to know today.
1. 🎯 Stop Measuring Your AI Product by Its Accuracy
Jeff Gothelf argues that AI product teams keep mistaking model outputs like “95% accuracy” for real Key Results, when OKRs should track how user behavior changes instead. He proposes a three-part structure. An Outcome KR watches observable user actions (”increase the share of meetings where the user shares the AI’s data without rewriting it from 40% to 65%”). A Calibration KR turns quality into checking behavior (”reduce sessions where the user opens the transcript to fact-check from 60% to 30%”). A Trust KR measures reliance by override rate (”keep manual overrides or rewrites of the AI’s analysis under 8%”). The core shift: trust shows up as people acting on the output without redoing it, not as a sentiment score. (Read More)
🫖 TEA For Thought: “Great post.”
2. ⚡ Two Harvard Dropouts Built a $1B Rival to Nvidia
Etched, founded in 2022 by Harvard dropouts and Thiel Fellows Gavin Uberti (CEO) and Robert Wachen (president), has hit a $5 billion valuation after raising $800 million total, including a $500 million round led by Stripes in December 2025. The company has booked $1 billion in contracts for “frontier inference clusters,” complete systems that pair its chips with custom racks and software. Its transformer-specialized inference chip, manufactured by TSMC earlier in 2026, is pitched as faster, cheaper, and more power-efficient than Nvidia’s, and is now in customer testing. Backers include Jane Street, Two Sigma, Peter Thiel, Andrej Karpathy, Geoffrey Hinton, and Fei-Fei Li. (Read More)
🫖 TEA For Thought: “I’ve met this group of Harvard-dropout founders at Etched. I was amazed then, and now, just wow.”
3. 💼 The AI Jobs Debate Just Got Messier
A new report from Ramp and Revelio Labs, covering nearly 22,000 companies, complicates the AI job-loss story: firms investing heavily in AI are growing headcount faster, not shrinking it. High-intensity adopters (about $30 per employee per month on AI) saw 10.2% headcount growth, and entry-level roles rose 12% at those firms. That cuts against gloomier data: roughly 90,000 job cuts have been tied to AI through May 2026, up to 15% of U.S. jobs could be automated within five years, and Goldman Sachs estimates AI erased about 16,000 net jobs a month over the past year, hitting Gen Z and entry-level workers hardest. The catch: the growth data skews toward venture-backed tech firms that were already expanding. (Read More)
🫖 TEA For Thought: “I guess only time will tell. For now, I think there are two reasons AI-pilled companies are hiring: 1) they have money, fresh capital injections; 2) AI’s abilities aren’t solid enough yet, so humans in the loop will stay with us, for a while.”
4. 🧠 Meta Decoded Typing Straight From Brain Waves
Meta’s AI research team, working with the Basque Center on Cognition, Brain and Language, built Brain2Qwerty v2, a system that reconstructs typed text from brain activity using a non-invasive MEG (magnetoencephalography) helmet, no surgery required. The model was trained on roughly 22,000 sentences from nine volunteers, each recorded for 10 hours while typing. It reaches 61% word accuracy overall and 78% for the best participant, where more than half of sentences decode with one word error or less. That is a leap from the 8% accuracy of prior non-invasive methods, and approaches results once exclusive to brain-surgery implants. Accuracy improved log-linearly with data, hinting the gap could close with scale, and could one day help people who have lost the ability to communicate. (Read More)
🫖 TEA For Thought: “How promising. Tech that changes life, literally.”
5. 🔓 An Apple Supplier Leak Exposed the iPhone 18 Pro
A ransomware group that stole data from Tata Electronics, an Indian Apple supplier, posted files on the dark web exposing the iPhone 18 Pro’s supply chain. At least six files map hundreds of components to the specific companies that make them, from main-board chips to battery and camera parts, revealing where Apple relies on a single supplier and where it holds leverage. Other images show iPhones in drop tests at a Tata plant in early 2026, depicting a slab-shaped grey handset with three rear cameras. Several files carried Apple “confidential” watermarks and iPhone 18 Pro code-names. Tata has restricted access to sensitive systems and hired a global consultant to run a forensic audit. (Read More)
🫖 TEA For Thought: “Here’s the thing about cybersecurity: it’s not just the company itself that needs to be protected. All the third parties that have access to the data are the greatest point of failure unless they’re strengthened too. That’s why limiting the frontier model to a few is just a patch, not a cure.”
🛠️ Skill of the Day
The Pre-Mortem: Find the reasons your plan fails before you commit to it.
You are my pre-mortem facilitator. We are about to [DECISION OR PROJECT, e.g. "launch a paid newsletter tier next month"].
Imagine it is [TIMEFRAME LATER, e.g. "six months from now"] and this has clearly failed. Do not reassure me.
1. Write a short, vivid story of exactly how it failed, in past tense, as if it already happened.
2. List the 8 most likely causes of that failure, ranked most to least probable. Separate the causes I control from the ones I do not.
3. For each of the top 3 causes, give me one concrete, cheap action I could take NOW to prevent it or catch it early.
4. Flag any single point of failure where one thing breaking sinks the whole plan.
Here is the context you need: [PASTE YOUR PLAN, GOAL, ASSUMPTIONS, DEADLINE, AND WHAT SUCCESS LOOKS LIKE].
Paste into ChatGPT, Claude, or your tool of choice. Replace the bracketed bits with your own. It surfaces the risks while you can still act on them.
TEAHEE Moment
Stay sharp, stay informed. See you tomorrow.
If you enjoyed this TEA, follow along on social for more:
Twitter/X