Levie shares detailed Box AI Complex Work Eval results: Fable vs Opus 4.8 across real enterprise knowledge work problems. Fable's main differentiators: it doesn't take shortcuts on complex reasoning, gets multi-step calculations right, and is significantly more consistent across runs.

Key results by industry: Media & Entertainment 78% vs 61%, Technology 81% vs 73%, Financial Services 89% vs 83%, Healthcare 66% vs 60%.

Specific examples: Legal M&A due diligence—Fable correctly identified joint-ownership clause violating exclusivity (100% vs 78%). Healthcare radiology error audit—Fable correctly concluded no Grade 3 errors existed while Opus prematurely escalated (63% vs 41%). Media genre profitability—Fable recognized a 20% Argentine tax was already embedded; Opus double-applied it (74% vs negative). Financial Services 5-year debt projection—Fable applied interest to opening balances correctly; Opus applied it to total facility (83% vs 62%). SaaS feature valuation—Fable exact arithmetic, Opus got multiple calculations wrong (100% vs 74%). 177 likes.

Thibault Sottiaux (OpenAI) confirms they saw a strong spike in growth of token consumption for Codex over the last 48 hours—unusual when they didn't launch anything. 2460 likes.

This is the competitive response effect: Anthropic's Fable launch drove users to push all AI coding tools harder, and Codex usage surged as a result. The rising tide lifts all boats.

Claude's 'Problem Solvers' series profiles Michael Truell, co-founder of Cursor: fell in love with coding at 12, went from 15 people to 700 in two years, and now over 60% of the Fortune 500 build with its AI coding platform. 5258 likes.

The speed of scaling is remarkable—15 to 700 in 24 months is a 46x headcount growth while maintaining product quality.

trq212 made a video about how he used Fable to edit Fable's own launch video. It wrote code and tool calls to use transcription services, ffmpeg, color grading, the Figma MCP, make a Remotion UI and render it. He didn't touch a video editor. 6036 likes—the highest engagement tweet of the day.

This is the meta moment: an AI model editing its own launch video using code, not a GUI video editor. The tool chain itself became the agent's workspace.

Josh Woodward confirms Gemini is experiencing a global outage. 'We're on it and will get everything back up ASAP.' 1553 likes. Later confirmed everything was restored.

A reminder that even Google-scale infrastructure can go down—and the timing, right after Fable's launch, amplified the competitive optics.

Zara Yang's sharp observation: 'It seems like most startups in San Francisco are selling products to each other.' When she asks founders who their target audience is, 90% of the time it's 'engineering and product teams, AI-native startups.' The same small group being bombarded with a million products, while very few people are building for the 99% of the world. 203 likes.

This echoes a growing concern: the SF AI bubble is creating products for SF AI people, not for the vast majority of potential users.

Zara Yang on cross-functional agents: teams should build agents/skills for their cross-functional partners. Example: a design team builds a design agent for the marketing team, trained on brand guidelines and patterns, so marketing can produce on-brand assets without bugging designers. The marketing team couldn't have built this themselves—it requires designer expertise and context. This moves teams toward organization by 'loops' rather than 'functions'. 50 likes.

Peter Yang's manifesto: give yourself permission to build. The traditional career ladder pushes everyone to become a leader, but some just want to be builders. As you climb, you're expected to fill time with product reviews, cross-functional alignment, managing up. Companies are now rewarding builders and ICs more than ever, but becoming a good builder takes reps—hard to do in back-to-back meetings. 52 likes.

Matt Turck's annual VC grind satire: '2026 is a BRUTAL grind in VC. You start in Davos, freeze in Aspen, hit Upfront, survive Milken, then straight to Paris for the French Open. Briefly back in NYC for the Knicks. Then total blur: SuperReturn in Berlin, Founders Forum in London, World Cup, Paris for Raise AI, Idaho for Sun Valley, quick respite in Mykonos, Goldman tech gauntlet, Slush in Finland, NeurIPS in Sydney... and boom, a productive year of thought leadership and adding value is over, and you're a wreck.' 488 likes.

Dan Shipper on AI and reshoring: he predicted on Lenny's podcast that higher productivity from each employee with AI makes it appealing to reshore certain jobs back to the US to be closer to customers. 164 likes.

This is an under-discussed second-order effect: AI doesn't just automate offshoring, it reverses it by making the location premium matter more than the cost arbitrage.

Garry Tan recommends Nessie: the best way to get all your existing context, memory and history from ChatGPT, Perplexity, and Gemini into all the other places you have memory, and also get it into OpenClaw/Hermes Agent. Their OpenClaw and MCP servers are ace. 269 likes.

Memory portability is becoming a real category: as users accumulate context across multiple AI platforms, the ability to move and unify that context becomes increasingly valuable.

Mike Krieger (Anthropic Labs, Instagram co-founder) talks with Dan Shipper about using Fable beyond day one. Key themes: (1) Fable feels like a teammate you delegate work to, not a tool you direct. He'll say goodnight to Claude, set up a complex task, and wake up to it done by 2am. (2) The model has 'judgment' — it pushes back on code review feedback instead of blindly accepting it, remembers feature flags you haven't turned on, and understands system-wide implications. (3) He built a personal media tracker app over a weekend where the app can modify itself — agent-native architecture taken to its logical extreme. (4) Dynamic Workflows are a game-changer: he ported a complex Python codebase to TypeScript/Bun over a weekend using a multi-step workflow with incremental testing and adversarial verification. (5) On whether software engineering is over: 'It's dramatically changed, but the craft of what needs do you have and is it actually good is still a very human endeavor.' Engineers feel both loss and excitement simultaneously. (6) Verification is critical: screenshots/video for every PR, FFmpeg scrubbing for animations, mock backends for complex systems. (7) Chat is not the only interface — mobile, progressive disclosure, and multiplayer are the next frontiers.