Good morning. Before you type anything spicy into your work laptop today, maybe look directly into the webcam and wave. Corporate America has entered its “we need to train AI on how people work” era. Meta took that literally, launching an employee tracking program to capture prompts, screen activity, and private conversations. Employees hated it. Then all the sensitive data Meta gathered leaked internally, forcing the company to pause the surveillance experiment. Funny how often the panopticon trips over its own power cord.
Anthropic is turning Slack into an orchestrator for AI agents. Teams can now add Claude as a team member in Slack channels, tagging it directly in threads and assigning it tasks. Claude breaks the job into steps and returns results right in the conversation. Anthropic says it uses Claude Tag to chase product metrics, handle support tickets, and find root causes of tricky bugs. The move makes AI delegation feel natural, since workflows start the same way you already bother your coworkers. Anthropic’s product team says their internal version of Claude Tag generates 65% of their new code. Soon, your entire AI agent stack might just be Claude getting tagged in Slack. (Anthropic)
Silicon Valley’s newest grift is called agent loops. Instead of giving an AI model one prompt and getting back an answer, agent loops have the model repeatedly guess, retry, and refine the result until another AI says it’s good enough. Anthropic demonstrated by having Claude build a retro style game. A single prompt took 20 minutes and cost $9. The agent loop version took six hours, cost $200, and mostly improved because Claude got more attempts. Silicon Valley is calling this innovation. Everyone else calls it charging 22x more for software that takes longer and tries harder. Only in AI can guessing repeatedly become a premium feature. (The Information)
Back to back meetings are where follow ups go to die.
Granola captures your notes, cleans them up, and turns the mess into clear summaries and action items so you can move things forward.
No awkward meeting bot. Just effortless notes that help you remember what mattered even when your calendar is packed.
👽 welcome to the future
A BCI startup wants to treat brain cancer like an electrical problem. Coherence Neuro temporarily placed a coin-sized implant in three patients during surgery to remove brain tumors. This was just an early safety test, but the company’s long term goal is bigger. Stanford researchers recently discovered aggressive brain tumors can hijack healthy neurons, tapping into brain signals to accelerate their growth. Mild electrical stimulation might disrupt this cancer growth. Right now, glioblastoma patients rely on MRIs every two or three months, leaving doctors fighting a deadly disease with snapshots. Coherence’s implant aims to continuously monitor tumors, slow growth with electrical pulses, and instantly alert doctors to any changes. “Come back in three months” is a bold strategy when the tumor has WiFi. (Wired)
AI agents are about to need passports. The Linux Foundation is launching Agent Name Service, an open standard for proving which company an AI agent represents online. It builds on DNS, the same internet plumbing browsers use to find websites. Once agents start touching customer data, payment systems, or payroll, companies will need proof of who sent them. The World Economic Forum says 82% of executives plan to adopt AI agents within a few years, meaning companies are about to have a lot of software walking around with permission slips. ANS can help verify who deployed the agent, but it can’t prove the agent has good judgment. (Silicon Angle)
AI companies want to turn feelings into data. Emotion AI is software that reads faces, voices, posture, and behavior to guess what people feel. It’s already showing up in call centers, where AI detects when customers sound frustrated and prompts agents to respond with more empathy. The next version wants more context, because a smile can mean someone is happy or one email away from quitting society. Companies sell it as AI that understands people better, but the same tools give workplaces and platforms a cleaner way to monitor them. Pretty soon, looking tired in a meeting might require a privacy policy. (IEEE Spectrum)
fun stats
👓 $299. Price of Meta’s new AI smart glasses. The Meta Glasses are about $80 cheaper than the older Ray-Ban Meta line and let you have Kylie Jenner’s voice as your AI assistant.
🇫🇷 77%. Mid-sized French companies using generative AI. Only 17% report any time savings. Très awkward.
☢️ 15 years. Walmart’s new long term nuclear power deal with Constellation Energy. The retailer locked in 176 megawatts, enough to power roughly 150,000 homes and prove nuclear demand goes beyond AI data centers.