Skip to content

Models that plan, call tools and finish multi-step tasks on their own — from Claude Code and Manus to agent frameworks and benchmarks.

1,790 picksRelated topicsMCP & tool useAI codingReasoning

Latest picks

1381–1400 of 1,790

May 9Saturday

Latent Space

Anthropic growing 10x/year while others lay off over 10% of staff

Anthropic is described as growing 10x annually and being valued at $1T-$1.2T, while the post cites layoffs of 40% at Block, 14% at Coinbase, and 20% at Cloudflare under AI-readiness framing.

Why it matters: Anthropic's annualized revenue is growing 80x while other labs cut over 10% of staff, showing money and talent concentrating at the top.

May 8Friday

AI HOT (Curated Pool)

Robotics Endgame: A Physical AGI Roadmap and LLM Analogy

The speaker presented a physical AGI roadmap with six named components: video world models, WAM, EgoScale, dexterity scaling laws, physical reinforcement learning, and DreamDojo; the snippet also mentions a 2016 OpenAI DGX-1 signing story with Jensen and Elon.

Why it matters: The speaker splits physical AGI into six modules and maps robot training onto the stages of large language model development, giving a shared framework for humanoid general intelligence.

Hacker News front page

Show HN: Git for AI Agents

regent-vcs released the open-source re_gent project for AI-agent version control, currently supporting Claude Code, with workflows for tracking why an agent changed files, rewinding sessions, and bisecting agent actions; the post does not disclose the license, storage format, or installation details.

Why it matters: regent-vcs's open-source re_gent adds Git-like version control to Claude Code, recording why the AI changed files, supporting conversation rollback, and using bisection to find which step caused a problem.

AI HOT (Curated Pool)

Running Codex Safely at OpenAI

OpenAI runs Codex with four safeguards: sandbox isolation, human approval, strict network policies, and native agent telemetry; the post does not disclose evaluation metrics, incident rates, or enterprise deployment requirements.

Why it matters: OpenAI published its internal Codex deployment security plan, splitting sandbox isolation, human approval, network allowlists and agent logs into four layers that others can adapt when setting code agent permission boundaries.

Alibaba Technology · WeChat

The AI-Native Era: Where R&D Organizations Go Next

Xu Xiaobin cites internal interviews showing that engineers who use AI heavily cut coding time from 30% to 5%, raised Agent conversation time from 5% to 60%, and increased end-to-end delivery efficiency by 2 to 3 times, while pure coding efficiency rose 10 times.

Why it matters: Alibaba's Xu Xiaobin says coding dropped from 30% to 5% of an engineer's time, with 60% now spent talking to agents.

Synced · WeChat

ICLR 2026: NVIDIA and Purdue Use an Agentic Loop for Text-to-3D Scene Generation

NVIDIA Cosmos Lab and Purdue University proposed Scenethesis, a language-and-vision agentic framework for text-to-3D scene generation that uses visual grounding, SDF-based physical constraints, and a judge module; experiments report about 72% first-pass success, 91% after self-checking, and collision rate reduction from 6.1% to 0.8%.

Why it matters: A concrete agent pipeline for 3D scene generation, with SDF physics constraints and a judge module that checks and corrects its own output.

QbitAI · WeChat

OpenAI releases three realtime voice models for reasoning, translation, and transcription

OpenAI launched GPT-Realtime-2, GPT-Realtime-Translate, and GPT-Realtime-Whisper as API models, covering 128K-context voice reasoning, streaming translation from more than 70 input languages into 13 output languages, and realtime transcription priced at $0.017 per minute.

Why it matters: OpenAI shipped three realtime voice models covering listen-while-reasoning, streaming translation across 70-plus languages, and transcription at $0.017 per minute, resetting both the reasoning ceiling and the cost floor for voice apps.

QbitAI · WeChat

All Labs Watch ByteDance, Everyone Praises DeepSeek: A U.S. Researcher’s 36-Hour China AI Trip

Ai2 researcher Nathan Lambert visited Zhipu, Moonshot AI, Tsinghua, Meituan, Xiaomi, and 01.AI within 36 hours, and said Chinese labs closely watch ByteDance and respect DeepSeek, while student participation in core work, open source habits, and in-house control of the technical stack mark key differences.

Why it matters: Ai2 researcher Nathan Lambert spent 36 hours visiting Zhipu, Moonshot AI, Tsinghua, Meituan, Xiaomi and 01.AI, and recorded how Chinese labs watch ByteDance closely and respect DeepSeek.

AI HOT (Curated Pool)

China releases L1-L4 AI terminal intelligence standards covering 7 device categories

MIIT and other agencies released AI terminal intelligence standards covering 7 device categories. The framework uses a “2+N” structure with L1 response, L2 tool, L3 assistance, and L4 collaboration; L4 details come later. The post does not disclose concrete test metrics.

Why it matters: MIIT and other agencies issued a national standard grading AI capability in phones, computers, glasses and four other device types from L1 to L4, with Xiaomi, Huawei and others helping draft it.

Ruan YiFeng's Weblog

Technology Enthusiast Weekly Issue 395: The Third Way of Software Development

Ruanyifeng Weekly issue 395 frames AI-assisted coding as a “mystery house” style of software development and cites HN SOTA, which ranks model popularity by scanning 200 top Hacker News topics each day and their programming or AI discussions.

Why it matters: In his weekly, Ruan Yifeng frames AI-assisted programming as mystery-house development, sets it beside the cathedral and bazaar models of open source, and argues it may become the default for individuals and small teams.

AI HOT (Curated Pool)

OpenAI launches official openai-cli for terminal API calls

OpenAI open-sourced openai-cli for direct API calls from the terminal. The Apache 2.0 tool installs via Homebrew or Go and covers Responses API, structured output, image editing, transcription, and key config. The key detail is Agent workflows using cloud tools like web search and code interpreter.

Why it matters: OpenAI open-sourced openai-cli, letting developers call the API from the terminal and chain cloud tools like web search and the code interpreter, skipping the SDK code they used to write.

Bloomberg Technology

Cloudflare to Cut One-Fifth of Workers in Move to AI-First Model

Cloudflare plans to cut over 1,100 jobs globally, about one-fifth of its workforce. The cuts are tied to an agentic AI-first operating model; the post does not disclose roles, timing, or cost targets.

Why it matters: Cloudflare plans to cut more than 1,100 roles, a fifth of its staff, and ties the move to shifting to an agentic AI-first operating model where AI agents take over parts of business processes.

AI HOT (Curated Pool)

Codex Plugin Now Supports Parallel Runs Across Chrome Tabs

OpenAI says Codex now runs in Chrome on macOS and Windows. The plugin works across tabs in the background without taking browser control; the post does not disclose version, concurrency limits, or enterprise policy.

Why it matters: OpenAI's Codex plugin now runs web and app tasks in parallel across background Chrome tabs on macOS and Windows without taking over the user's browser.

NVIDIA Blog

Powering the Next American Century: Chris Wright and NVIDIA’s Ian Buck on Genesis Mission

The U.S. DOE and NVIDIA are building two AI supercomputers at Argonne; Equinox uses 10,000 Grace Blackwell GPUs. Solstice will use 100,000 Vera Rubin GPUs, which Buck said reach 5,000 exaflops. The key bottleneck is grid work: Wright said AI can cut interconnection studies from years to weeks or hours.

Why it matters: The US Department of Energy and NVIDIA are deploying two AI supercomputers at Argonne National Laboratory, scaling from 10,000 Grace Blackwell chips to 100,000 Vera Rubin GPUs.

AI HOT (Curated Pool)

Agent Pull Requests Are Everywhere: How to Review Them

GitHub published a guide for reviewing pull requests generated by AI agents. The snippet lists 3 focus areas: code changes, logic or security bugs, and pre-merge technical debt. The key issue is a review process before automated commits reach production.

Why it matters: A GitHub blog post lays out a three-step method for reviewing AI agent pull requests, from reading the diff to hunting logic and security bugs to clearing tech debt before merge.

AI HOT (Curated Pool)

Work with Claude across Excel, PowerPoint, Word, and Outlook

Claude now connects to four Microsoft apps: Excel, PowerPoint, Word, and Outlook. Excel, PowerPoint, and Word are generally available; Outlook is in public beta. Admins can deploy via Microsoft admin center and monitor with OpenTelemetry.

Why it matters: Claude now works inside Excel, PowerPoint, Word and Outlook, with the first three fully open and Outlook in public beta, keeping context across apps in one conversation.

AI HOT (Curated Pool)

Perplexity launches Personal Computer app for Mac

Perplexity opened its Personal Computer Mac app to all users. It runs on any Mac and works across local files, native Mac apps, the web, and Perplexity secure servers. The post does not disclose pricing, permission boundaries, or task success rates.

Why it matters: Perplexity turned its computer-controlling AI assistant into a Mac app open to everyone, no invite code needed, running tasks across local files, native apps and the web.

May 7Thursday

AI HOT (Curated Pool)

Trillion-parameter instruction model Ling-2.6-1T released

inclusionAI says Ling-2.6-1T is now live on OpenRouter. The trillion-parameter instruction model uses “fast thinking” and claims top AIME26 and SWE-bench Verified results with about 75% lower cost. The post does not disclose pricing, context length, or full benchmark scores.

Why it matters: inclusionAI put its trillion-parameter instruction model Ling-2.6-1T on OpenRouter, pitched at fast thinking and claiming inference costs about 75% lower than comparable models for advanced coding and agent workflows.

Hacker News front page

AlphaEvolve: Gemini-powered coding agent scaling impact across fields

Google DeepMind describes AlphaEvolve as a Gemini-powered coding agent; the body is only an RSS snippet. The title discloses coding-agent scope and cross-field impact, but the post does not disclose model version, benchmarks, or deployments.

Why it matters: DeepMind explains how AlphaEvolve, its Gemini-powered coding agent, is being applied across several fields, giving readers a look at one approach to cross-domain use of coding agents.

AI HOT (Curated Pool)

Apify mcpc and x402 Give AI Agents an Auto-Payment Wallet

Apify mcpc integrates the x402 payment protocol, letting AI agents auto-sign payments on HTTP 402. x402 compresses paid API settlement into one HTTP round trip plus a signature; mcpc supports Claude Code and USDC-funded wallets. The key point is machine settlement for paid tool calls, not the wallet label.

Why it matters: Apify's mcpc wires the x402 payment protocol into a general MCP client, so agents hitting an HTTP 402 can sign and settle automatically, with no signup or card.