·
Finding signal on Twitter is more difficult than it used to be. We curate the best tweets on topics like AI, startups, and product development every weekday so you can focus on what matters.
PLUS: Coding loops are burning your money
OpenAI just published their full plan for achieving the mission and what comes next for the company.
Meta AI grew 2.5x in two months but only 4.5% of users stick around after 30 days, making it growth theater.
AI agents struggle with biology databases because they're built for humans, not machines. Infrastructure matters more than models.
NotebookLM now has agentic chat and advanced reasoning powered by Gemini 3.5, rolling out to Ultra subscribers first.
Data is the real driver of AI progress. Catching up is easy when you can distill from APIs but not hyper-parameters.
Apple's latest keynote accidentally outlined an intent-centric OS where you don't need apps. AI constructs interfaces on demand.
v0 Max now runs on Claude Opus 4.8 for better code generation and design.
Google ordered 3 million TPUs from Intel. The government's 10% stake is paying off hard.
Chrome DevTools now has AI assistance with agent walkthrough and copy-to-coding-agent features built in.
Running AI coding loops burns outrageous money unless you're deeply technical. Work in small chunks instead.
Benedict Evans says AI feels like the internet in 1997. Coding agents found product-market fit, but vertical products matter most.
Using ChatGPT doesn't make you AI native. That's like having a website and calling yourself a tech company.
Bending Spoons went from 4 months of runway to $1.3B annual revenue and just filed to go public.
Runway's Aleph 2.0 lets you upload a video and reformat it for every aspect ratio by filling in missing scenes.
Every transportation tech expands our radius but flying cars eliminate the speed-freedom trade-off. The time is finally right.
Stepful raised $55M to train healthcare workers debt-free while working full-time. Over 32,000 graduates so far.
Both Anthropic and OpenAI mention slowing AI development in their latest posts but only if coordinated globally somehow.
When electricity arrived, factories just swapped the steam engine for electric and got nothing. Real gains came from redesigning everything.
Nebius is going head-to-head with hyperscalers. The biggest threat is extreme consolidation into three to five closed empires.
There aren't infinite commercializable problems in a domain. Running loops 24/7 ships green diffs into the void unless coupled with distribution.
Ramp's newsletter works because it's built on proprietary spend data from 50,000+ companies. You'd read it even if they sold nothing.
Generalist says robotics models are crossing into GPT-3-level commercial viability. A crossover point for reliability and speed.
Brilliant launched Koji, an AI tutor that uses Socratic method instead of handing out answers. Making kids think, not dumber.
Every LLM is actually a compression machine disguised as next token prediction. Better compression means better reasoning.