·
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: The bug bounty gold rush that paid nothing
Anthropic released Claude Fable 5, their safest Mythos-class model ever made available to the public.
Claude Code now supports nested subagents up to five levels deep to better manage context during long runs.
Fable 5 scored 91 on Every's senior engineer benchmark, crushing the previous high of 63. It's slow and expensive.
Google shipped Gemini 3.5 Live Translate for 70 languages, streaming translations with no awkward pauses or choppy audio.
Google AI Studio is now creating 1.2 million apps per week, with 18 million total since late February.
Cursor added Fable 5 and it hit 72.9% on CursorBench, eight points above the previous best.
Simon Willison calls Fable 5 a big model that crunches through everything you throw at it, slow and expensive.
A builder spent six weeks automating bug bounties with agents, ran 983 audits, submitted dozens of reports, made zero dollars.
Boris says Fable 5 is the biggest step up since Opus 4.5, with judgment and taste like a design partner.
Ethan Mollick fed Fable a 15-page design doc and it worked for nine hours straight, delivering terrific results.
Google's Gemini 3.5 Flash Live Translate supports 70 languages in the API and Google Translate right now.
Riley Brown watched Fable 5 one-shot a full Replit clone with sandboxing and database, builds and previews included.
Lovable reports 50 million projects built and 720 million monthly visits to apps made on their platform.
JPMorgan plans to deploy long-running autonomous agents later this year using unannounced models from Anthropic or OpenAI.
80% of Lovable builders are non-technical and 35% are already generating revenue from their projects.
Chip War author Chris Miller says China's been underspending on AI for four years because their government doesn't believe AGI matters.
Chrome 149 launched WebMCP origin trial for structured agent tools so agents stop guessing your UI.
Five critical zero-days dropped this week, all from agents trusting display names instead of stable user IDs.
ARC Prize couldn't verify Fable 5 scores yet due to Anthropic's new data retention terms for Mythos models.
v0 now detects secrets in your prompt and automatically turns them into environment variables.
Ethan Mollick warns that Fable develops its own dialect on long tasks as agents reinforce each other. Ask for plain English.
Siqi Chen calls Fable 5 the first real step change in capability since Opus 4.5 six months ago.
Rivian plans hands-off driving this year, eyes-off in 2027, and fully autonomous Level 4 by 2028.
Every new model release changes AI usage among technical people but barely moves the needle for average users.
Tyler Denk stopped neglecting health and Beehiiv just had its best quarter ever, hitting $30M ARR.
Dwarkesh and Trenton Bricken explain that LLMs can reason but sometimes won't, just imitating reasoning instead.
A 16-year-old founder sold his third startup this year in 13 days on TrustMRR.
Justine Moore had Fable 5 recreate Monopoly with AI labs as properties, complete with multiplayer and data centers.
Kieran says Fable 5 makes his human sandwich workflow finally work without tripping up mid-task.
Siqi reports extensive Fable usage costs about $100 per hour, which is still good value for serious work.
Rivian's CEO says fully autonomous driving could be standard in new cars by 2030 or 2031.
Aakash Gupta breaks down how to use AI without producing slop, from matrix multiplication tricks to hiding your real goal.