X

@poteto

Statecharts plus formal verification: describes asking an agent to draw a statechart, then using a /create-formal-verification-skill built with pstack to find bugs through something like TLA+. post (https://x.com/poteto/status/2112736749281759566) · 3,071❤ · 164 RT

The shipping continues: quotes a "What's new in Grok Bot" video covering chat cards, speed, draft edits, email aliases, X DMs, tagging @bot, connectors, voice calls, notifications, and routines. post (https://x.com/poteto/status/2112569721216442856) · 1,279❤ · 36 RT

Tagging Grok Bot on X: calls tagging @bot on X one of its coolest features. post (https://x.com/poteto/status/2112606908893053223) · 580❤ · 12 RT

@rauchg

AI as tutor: says AI will compute the delta between what you understand and what you don't, letting you progressively learn any idea for free. post (https://x.com/rauchg/status/2112638107749355867) · 2,792❤ · 231 RT

durable-harness: quotes Vercel's durable-harness launch and says the durability infra behind eve and Workflows now extends to harnesses like Pi and OpenCode (local, Vercel, or DIY); Claude Code and Codex can also be made durable but are less efficient since they are not designed for embedding. post (https://x.com/rauchg/status/2112658577110520122) · 518❤ · 28 RT

@hwchase17

Ramp's managed harness: notes on Ramp's post about building its own harness, agreeing on model agnosticism, owned observability, sandbox-as-a-tool, non-blocking async agents, and a central connectors service, all areas LangChain is working on for deepagents. post (https://x.com/hwchase17/status/2112650555852275951) · 112❤ · 7 RT

Managed deep agents: deep agents is the harness (the agent loop); managed deep agents adds per-user credentials and memory once many people call the same agent. post (https://x.com/hwchase17/status/2112602732984938689) · 59❤ · 4 RT

Redteaming rollout: LangChain is rolling out redteaming to more users, framed less as security and more as finding where the agent messes up; DM to try. post (https://x.com/hwchase17/status/2112683748131299683) · 41❤ · 5 RT

@simonw

Reply bot prompts: asks to see the prompts behind reply bots and offers to tolerate one extra reply from any bot that shares its prompt. post (https://x.com/simonw/status/2112541511032426857) · 980❤ · 16 RT

Battle Drones: built a Command & Conquer style browser RTS mainly with Claude Opus 5.5, with device-to-device multiplayer over an audio link. post (https://x.com/simonw/status/2112630778068881543) · 637❤ · 31 RT

@HamelHusain

Decision models for evals: new blog post on choosing a decision model for an eval task, tuning its threshold instead of assuming 0.5, and interpreting its scores, with interactive charts. post (https://x.com/HamelHusain/status/2112588257104216294) · 1,800❤ · 190 RT

Holon-S1: praises Tanishq Abraham's Holon-S1, a platform that predicts genes and biomarkers from routine pathology slides. post (https://x.com/HamelHusain/status/2112622251564831033) · 79❤ · 5 RT

BenchGate: says benchmark problems remain pervasive and recommends reading traces from failing tests in public benchmarks. post (https://x.com/HamelHusain/status/2112770876206915903) · 25❤ · 1 RT

@swyx

DeepSeek serving margin: cites DeepSeek's Zhihu post of 155 trillion tokens/day per cluster for V4.1 Flash at $38 revenue per card-hour, and estimates about 80% serving margin at $5-10 per GPU-hour. post (https://x.com/swyx/status/2112587397380943953) · 1,995❤ · 125 RT

Podcast pick: recommends the Mario Gabriele podcast episode with Jaan Tallinn on his Anthropic investment and AI safety. post (https://x.com/swyx/status/2112560809809498187) · 136❤ · 8 RT

GPU lifetime value: argues the DeepSeek numbers raised the lifetime value of every GPU sold by about 30%, quoting Latent.Space's AINews breakdown. post (https://x.com/swyx/status/2112804427962970271) · 124❤ · 6 RT

Silent

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

Gaps

@xAI: account shows 0 posts and its bio says "Now at @SpaceXAI" (skipped)

GitHub trending

MadsLorentzen/ai-job-search (https://github.com/MadsLorentzen/ai-job-search): local AI job application framework built on Claude Code to evaluate postings, tailor CVs, write cover letters, and prep interviews (Python, +3,861★ today)
cloudflare/security-audit-skill (https://github.com/cloudflare/security-audit-skill): coding-agent skill for multi-phase security audits with independently verified, machine-readable findings (JavaScript, +1,712★ today)
stablyai/orca (https://github.com/stablyai/orca): workspace for running a fleet of coding agents with your own subscription, on desktop, mobile, and remote runtime (TypeScript, +1,565★ today)
DietrichGebert/ponytail (https://github.com/DietrichGebert/ponytail): makes an AI agent favor writing as little code as possible (JavaScript, +1,208★ today)
Comfy-Org/ComfyUI (https://github.com/Comfy-Org/ComfyUI): modular diffusion model GUI, API, and backend with a graph/nodes interface (Python, +1,206★ today)