Claude Code now supports persistent messaging channels through Telegram, Discord, and iMessage. Users can text their AI agent from their phone and it responds with full access to local files and tools. Early adopters are already building always-on AI coworkers that check in via messaging apps, run background tasks, and push proactive updates — without anyone opening a browser tab.

Verve Therapeutics published Phase 1 results showing prime editing successfully modified the PCSK9 gene in 12 patients with familial hypercholesterolemia. LDL cholesterol dropped 55% on average with a single infusion. No off-target edits were detected at the 6-month follow-up — a critical safety milestone that previous gene therapies have struggled to clear.

Andrew Huberman reviewed 14 meta-analyses on sleep quality interventions. His conclusion: morning sunlight exposure (10–30 min), consistent wake time (even on weekends), and magnesium threonate before bed are the only three with replicated, large-effect evidence. Everything else — blue light glasses, sleep trackers, weighted blankets, melatonin — showed weak or inconsistent effects across populations.

A new NBER paper analyzing 24 months of Kalshi CPI contracts found that prediction market median forecasts outperformed the Wall Street consensus on 85% of prints, with a mean absolute error 40% lower than professional forecasters. The authors attribute the gap to real financial stakes removing the career-risk bias that pushes institutional economists toward consensus clustering.

Of 240 startups in YC's Summer 2026 batch, 144 describe themselves as AI-native. The largest clusters: AI agents for specific industries (42), AI-powered dev tools (28), and AI applied to healthcare and biotech (23). Notably absent: general-purpose chatbots. The era of "ChatGPT wrapper" startups appears to be over — YC partners are now explicitly filtering for vertical specificity and defensible data moats.