Vercel's Next.js 16.3 Cuts Prefetch Requests by 45%

Vercel's Next.js 16.3 delivers major performance gains — 45% fewer prefetch requests, 17% less static assets, and 2x faster path metadata — with a simple upgrade path that cuts compute and data transfer costs.
Stripe unveils a roadmap for programmable money movement across Treasury, Payouts (160+ countries), Capital, and Stablecoins, expanding its financial infrastructure.
Shopify Q2 shows strong momentum: $116B merchant sales (+32% YoY), $3.6B revenue (+34%), and $654M free cash flow, with GMV growing over 30% for five straight quarters.
Google DeepMind is set for a leadership transition: the author of a widely shared post (retweeted by @jack) will become DeepMind Chair and Alphabet Chief Scientist, with Koray taking over GDM leadership.
New AI infrastructure entrants include Fly's Serverless Agents (~90-second agent deployment) and Pokee-Isaac 28B, claimed as the first true 10M-token-context agentic model on a single GPU (unverified).

Next.js 16.3 from Vercel significantly improves service efficiency and build speed: it cuts 45% of prefetch requests, 17% of static assets, and makes path metadata 2x faster. The upgrade path is simple, and applying it on Vercel substantially reduces compute and data transfer costs at scale. — via 1
Vercel is expanding into the backend: FactoryAI runs its API service on Vercel's Fluid Compute, handling billions of requests per month. This demonstrates that Vercel can serve as the "Vercel for the backend." — via 1
DeepSeek v4 Flash cost optimization: using @aisdk's one-liner saves over 90% on AI Gateway tokens. Separately, Novita offers Vercel Pro users a 90% discount on DeepSeek v4 Flash through August 11. — via 1
Fly's Serverless Agents lets developers deploy an agent worker with API, tool integrations, and background runs in ~90 seconds, eliminating the need to build agent infrastructure. — via 1
Pokee-Isaac 28B is promoted as a frontier-level agentic model with true 10M-token context and single-GPU deployment. These claims are unverified and should be treated with caution. — via 1

Stripe will release a broad set of programmable money movement capabilities across Treasury, Payouts (covering 160+ countries), Capital, and Stablecoins. A full roadmap is now public, opening new automation possibilities for financial flows. — via 1
Block launched an investor relations bot that answers natural-language questions using public materials (shareholder letters, 10-K/10-Q, investor day content). Each answer includes source citations, supports multiple languages, and only draws on public IR materials. — via 1
Strike is deploying a frontier AI agent red team to continuously challenge its systems, because AI has made vulnerability discovery dramatically cheaper and enables attackers to investigate codebases like expert teams. Engineers validate attack paths and handle fixes. — via 1
Telegram's brief removal from the App Store highlights a systemic risk: attackers used stealthily tampered content to report the app and extort the company. This serves as a warning for other mobile app developers facing similar takedowns. — via 1

Shopify Q2 results are strong: merchant sales reached $116B (+32% YoY), revenue $3.6B (+34%), and free cash flow $654M. GMV has now grown over 30% for five consecutive quarters; tobi called it a "Good quarter." — via 1
SPC announced its fourth fund at $575M, continuing its mission to help founders find their life's work. Amjad Masad noted the high founder talent density in the SPC incubator. — via 1
SparkToro V3 launched after 2.5 years, its first new version since 2020 (third overall). It brings new features, charts, and a comprehensive upgrade, integrated with a newly announced API and MCP. — via 1
Google DeepMind leadership change: a widely shared post (retweeted by @jack) says the author will become DeepMind Chair and Alphabet Chief Scientist, focusing on long-term strategy and scientific breakthroughs, including using Isomorphic to treat diseases. Koray will take over GDM leadership. — via 1

Perplexity Computer emphasizes verification for high-risk research: in financial queries, every number is traceable to its original source and the full calculation trail is displayed. — via 1
levelsio's self-quantified productivity analysis finds that personal-trainer sessions with high training intensity correlate with +20% productivity, even after accounting for training and shower time. Alcohol, long flights, illness, high-fat diet, and low protein correlate with lower productivity; most results are directional hypotheses, not statistically significant. — via 1
levelsio warns against AI batch refactoring: having AI do a large-scale "Gauntlet Loop" refactor at once results in messy, inefficient code and wastes about $500. He advocates step-by-step, feature/object-level control and calls such stunts more about attracting X traffic than maturity. — via 1

Discover