X

@karpathy

LLM capability funnel: revisits his April post, arguing the gap in understanding is widening into a funnel: ~6B people barely use LLMs, ~1-2B casual free-tier users, ~20M professionals using frontier models for math and code, ~5,000 with internal frontier access; drivers are problem ambition, jagged capability in verifiable domains, and access tier. post (https://x.com/karpathy/status/2109361546505966046) · 32,867❤ · 4,258 RT

@poteto

Rethinking clean code: argues maintainability-focused clean code matters less now that agents write the code, since readability has a performance cost; expects new principles of agent-friendly architecture. post (https://x.com/poteto/status/2109356992032817621) · 3,501❤ · 176 RT

Verification as the first agentic loop: recommends giving agents a way to run and check their own work like a user (CLI, API, UI, even TLA+), producing artifacts such as screenshots or video that prove correctness and are easier to review than the PR. post (https://x.com/poteto/status/2109436202554372476) · 2,058❤ · 88 RT

Desktop offline fix in 0.70.2: fix for the Grok Bot desktop client going to sleep and showing a machine as offline is released in 0.70.2 and rolling out. post (https://x.com/poteto/status/2109437986870100379) · 1,411❤ · 38 RT

@swyx

AIE NYC talk tips: reshares his talk-preparation tips thread for AI Engineer NYC speakers ahead of the speaker dinner. post (https://x.com/swyx/status/2109365419937239158) · 36❤ · 3 RT

@hwchase17

Future agent architecture slide: maps a slide from Lee Robinson's Stanford lecture on always-on agents to LangChain managed deepagents: durable workflows, sandboxes, memory and skills, schedules, connectors, channels, evals, observability; payments and identity still missing. post (https://x.com/hwchase17/status/2109307704346853833) · 334❤ · 37 RT

Always-on agents lecture slides: shares extracted slide screenshots from Lee Robinson's Stanford CS146S lecture on proactive agents. post (https://x.com/hwchase17/status/2109297234864971878) · 221❤ · 26 RT

Agent identity: asks whether internal agents doing proactive background work should keep acting on behalf of the user or get their own identity like coworkers. post (https://x.com/hwchase17/status/2109270222163066890) · 92❤ · 1 RT

@rauchg

Anatomy for learning: quotes Anatomy, a Claude-based tool that turns a one-line machine description into interactive 3D wireframes, calling it the future of learning. post (https://x.com/rauchg/status/2109297685392175488) · 2,815❤ · 105 RT

@HamelHusain

Excalidraw for slides: recommends Excalidraw with its MCP for building slides with animations and clean visuals. post (https://x.com/HamelHusain/status/2109365383086170602) · 360❤ · 9 RT

Which role wins: summarizes marketer, PM, and engineer claims to be the key role; expects strong people in each to stay valuable and none to dominate. post (https://x.com/HamelHusain/status/2109371836056350802) · 137❤ · 8 RT

Silent

@martinfowler, @cursor_ai, @simonw: no original posts in window

Gaps

@xAI: API not authorized to read this account's posts (skipped)

GitHub trending

morluto/rea (https://github.com/morluto/rea): reverse engineer anything with agents, from app behavior down to native binaries (TypeScript, +25,599★ today)
boykopovar/AnyPS5 (https://github.com/boykopovar/AnyPS5): tool for automatic PS5 executable porting to Linux and Windows (C++, +3,952★ today)
storytold/artcraft (https://github.com/storytold/artcraft): intentional crafting engine for artists, designers, and filmmakers (Rust, +2,957★ today)
cathrynlavery/diagram-design (https://github.com/cathrynlavery/diagram-design): editorial diagram design for coding agents, 44 diagram types as self-contained HTML + SVG (HTML, +1,083★ today)
bilawalsidhu/gods-eye-view (https://github.com/bilawalsidhu/gods-eye-view): browser spy-satellite simulator with real data, open-source spatial intelligence on a photorealistic 3D globe (JavaScript, +886★ today)