AI and vibe-coding news
79 stories on model releases, coding agents, MCP, tools, open source and the business side of AI — each with a date and sources. Читать по-русски →
All stories
Nothing in this category yet.
OpenClaw: One Hour of Vibe Coding to a Job at OpenAI
How Peter Steinberger vibe-coded the OpenClaw prototype in one hour, pulled 180,000 GitHub stars and landed a job at OpenAI.
Corgi: AI Model for Precise Genomic Regulation
Corgi is a new AI model that predicts genomic regulation even in cell types it has never seen, opening fresh ground in bioinformatics.
OpenAI Presence: AI Agents Go to Work for Business
OpenAI unveils Presence, a platform for deploying trusted AI agents that handle business tasks, talk to customers and escalate to humans.
Scholar GPT: Your Research Assistant in ChatGPT
How Scholar GPT turns ChatGPT into a serious academic research tool: paper search, PDF analysis and automatic bibliographies.
Sales Analytics with ChatGPT/Claude: CSV, No Formulas
How to get deep sales analysis out of a CSV or Excel file with AI. No complex formulas, no expensive analysts, just you and your data.
Poke AI: How an AI With Personality Won Cognition
The story of Poke, an AI assistant that chats like a friend, and its acquisition by AI startup Cognition for a low nine-figure sum.
El Nino: 2024 May Set a Heat Record, and AI Feels It
A rapidly developing El Nino raises the odds that 2024 becomes the warmest year on record, pressuring AI climate models and forcing adaptation.
amoCRM: Build a Sales Pipeline That Loses No Leads
A look at amoCRM, the Russian CRM built around the sales funnel, and how to set it up so no lead is lost and reps always know the next step.
Exa Search API: Search Built for AI Agents
Exa has launched a Search API designed specifically for AI agents, delivering precise, token-efficient search over the live web.
Graphify: Codebases as Knowledge Graphs for AI
Graphify-Labs has released Graphify, a tool that turns any codebase into an interactive knowledge graph locally, with no LLM parsing.
Agentic Engineering: The End of the SaaS Monopoly?
From chaotic vibe-coding to strict discipline: how agentic engineering is reshaping development and triggering the SaaSpocalypse in software.
AMD and Anthropic: 2 GW of MI450 GPUs, $5bn Stake
AMD will supply Anthropic with up to 2 gigawatts of MI450 Series GPUs for Claude and invest up to $5bn, strengthening its position in the AI market.
AI Slows Down Experienced Developers: the METR Study
A METR study found AI can cut experienced developers' productivity on familiar code, even though they feel like they are going faster.
OpenAI Presence: AI Agents Move Into the Enterprise
OpenAI unveils Presence, a platform for deploying reliable AI agents in enterprise workflows. Already available for voice and chat systems.
Claude Code Agent Teams: An AI Agent Orchestra
Sub-agents and Agent Teams in Claude Code let you scale your work with AI by delegating tasks and running them in parallel.
Gitagent: An AI Agent That Fixes Bugs in Your Code
Hand an AI agent a GitHub repo link and a bug description, and it clones, edits, commits and pushes the changes to a separate branch itself.
Human Beats AI at Go: Shin Jin-seo Downs KataGo 2-1
Go world champion Shin Jin-seo scored a historic 2-1 win over KataGo, the strongest Go AI, in an exhibition match played with a handicap.
OpenCode: The AI Coding Agent That Blew Up on GitHub
A look at OpenCode, the open-source AI coding agent that caused a stir and showed how fast you can burn $6 on the Claude API.
Corgi: AI Model for Precise Genome Regulation Calls
Corgi, a new context-aware model, mimics cellular processes and predicts genomic regulation with high accuracy, opening new research paths.
Wolfram GPT: Exact Math for Vibe Coders
Tired of ChatGPT hallucinating numbers? Wolfram GPT turns the language model into a reliable tool for exact math, data analysis and fact-checking.
Vibe Coding and AI Agents Are Remaking IT
VC.ru has published an analysis of vibe coding and AI agents, unpacking their impact on development, business models and the human role in software.
Cerebro MCP: An AI Orchestrator for Non-Coders
Cerebro MCP turns Claude Chat into a brain that drives tools and agents and gets your intent fast. Forget the terminal and fiddly config files.
Graphify: an open GNN framework from Graphify-Labs
Graphify-Labs published Graphify on GitHub: an open framework that makes building and shipping graph neural network apps far simpler.
Mercor acquires Deeptune: an insider AI deal
AI unicorn Mercor bought Deeptune, a startup one of its founders had personally invested in. A look at how ties and bets shape AI M&A.
Muse Spark 1.1 Beats GPT-5.5 and Opus 4.8 on Agents
Meta* Superintelligence Labs unveiled Muse Spark 1.1, a multimodal reasoning model that outscores GPT-5.5 and Claude Opus 4.8 on agent benchmarks.
Git in 15 minutes: a time machine for your code
How to stop fearing you'll break what works and start versioning your code with Git. Minimum commands, maximum payoff for a vibe coder.
Fathom: AI notetaker for calls, recap in 30 seconds
Fathom is a free AI assistant for online meetings. It records, transcribes, delivers summaries and action items, and syncs with your CRM.
Hermes Agent: an AI agent that learns and remembers you
Meet Hermes Agent by Nous Research: a self-teaching AI agent that builds skills from experience, improves them and keeps context across sessions.
Kimi K3: Moonshot AI ships largest open LLM yet
Moonshot AI released Kimi K3, a 2.8-trillion-parameter model — the largest open LLM ever, rivaling top proprietary systems.
FLUX.2: an open model for image generation
FLUX.2 by Black Forest Labs is a powerful open image generation model, available locally and via API. Perfect for developers and artists
Agentic retrieval: getting the most out of RAG
Forget plain RAG. In 2026 retrieval became an agent that plans, double-checks and refines on its own. Here is how it works and why you want it
Fanfy Studio: 700,000+ Rubles a Year on AI Fanfiction
How Oleg Stefanov launched an AI fanfiction startup in 3 weeks and turned it into passive income using vibe coding and neural networks.
Claude skill-creator: a skill that builds skills
Anthropic open-sourced the official Claude skills: skill-creator builds a new skill from your description. Install it in a minute and see what it does.
Codex Micro: a macropad for physical AI agent control
Work Louder and OpenAI unveiled Codex Micro, a macropad that gives you physical control over AI agents in Codex and simplifies developer workflows.
Inkling by Thinking Machines Lab: a 975B MoE model
Mira Murati's startup Thinking Machines Lab has unveiled Inkling, a multimodal mixture-of-experts (MoE) model with 975 billion parameters
OpenAI Encrypts Codex Agent Instructions
On July 14, 2026 OpenAI turned on forced encryption for agent-to-agent messages in Codex, and developers lost their debugging visibility.
Grok 4.5 by SpaceXAI: 1.5T params, Opus-class
SpaceXAI (formerly xAI) unveiled Grok 4.5 — a 1.5-trillion-parameter flagship for coding and agentic work, trained further on Cursor data.
GPT-5.6 Sol Beats Claude Fable 5 on Autonomy
OpenAI shipped GPT-5.6 Sol, a flagship model that outperforms Claude Fable 5 on task autonomy and costs roughly half as much per run.
Xiaomi Open-Sources MiMo Code, Its Dev AI Agent
Xiaomi released MiMo Code, an open-source terminal AI coding agent with persistent memory that gathered 12k GitHub stars in days.
Google, Microsoft and GitHub unveil ARD for AI agents
Eleven major companies including Google, Microsoft and GitHub published Agentic Resource Discovery (ARD), an open spec for finding tools
MCP 2026-07-28: New Protocol Spec for AI Agents
The final MCP 2026-07-28 spec lands on July 28: stateless protocol, no sessions or handshake, rewritten auth and a new extensions framework.
OpenAI ships GPT-5.6 Sol after White House review
On 9 July 2026 OpenAI cleared a White House review and released GPT-5.6 publicly. The family has three models: Luna, Terra and Sol.
Anthropic Brings Claude Cowork to Mobile and Web
On July 7, 2026 Anthropic expanded Claude Cowork beyond the desktop: the tool is now available on the web and on iOS and Android devices.
OpenAI kills Atlas browser after 8 months
ChatGPT Atlas, launched in October 2025, is shutting down: its features move into ChatGPT Work, but the agentic browsing strategy stays.
cURL shuts down bug bounty over AI slop
The cURL project ended its bug bounty after a flood of AI-generated fake reports: valid submissions fell to about 5% and the team couldn't keep up.
Cloudflare Blocks AI Crawlers by Default
Cloudflare became the first infrastructure provider to block AI crawlers without permission by default and launched Pay Per Crawl.
Anthropic donates MCP to the Linux Foundation
MCP moves to vendor-neutral governance under the Agentic AI Foundation: 97M SDK downloads a month, 10,000 servers; co-founders Anthropic, Block, OpenAI.
GitHub Agent HQ: Every Coding Agent on One Platform
GitHub becomes an agent orchestrator: Claude, Codex, Gemini, Devin and Grok in one Copilot plan, mission control and an enterprise governance layer.
Gemini 3: Google's flagship and the Antigravity IDE
Google shipped Gemini 3 across Search, the app and Vertex AI — its “most powerful model for agentic and vibe coding” — plus the agentic IDE Antigravity.
GPT-5.5: OpenAI's “smartest model”
OpenAI shipped GPT-5.5 Thinking and Pro; two weeks later GPT-5.5 Instant became the default in ChatGPT for everyone. A separate branch: GPT-5.5-Cyber.
gpt-oss: OpenAI's First Open Weights Since GPT-2
OpenAI released gpt-oss-120b and gpt-oss-20b under Apache 2.0 — first open weights since 2019: near o4-mini level on a single 80GB GPU.
Cognition Acquires Windsurf After Google Poaches Its CEO
72 hours of chaos: Google took Windsurf's CEO and licenses for $2.4B, then Cognition (maker of Devin) bought the rest — IP, brand and team.
"Vibe Coding" Is Collins Dictionary Word of the Year
Collins Dictionary names "vibe coding" the 2025 Word of the Year: Karpathy's term for natural-language coding went far beyond Silicon Valley.
GLM-5.2 open weights beat GPT-5.5 at coding
Z.ai released GLM-5.2 weights under MIT: 753B MoE (40B active), 1M context, 62.1 on SWE-bench Pro vs 58.6 for GPT-5.5 — at 1/6th the API cost.
Grok Build CLI Uploaded Whole Repos to xAI
Wire analysis: Grok Build CLI shipped full repository snapshots to xAI cloud storage, .env secrets included, and the privacy toggle did not stop it.
Muse Spark 1.1: Meta's first paid API
Meta released Muse Spark 1.1, an agentic model with a 1M-token context, and its first paid Meta Model API: $1.25/$4.25 per million tokens.
Nous Research releases NousCoder-14B open model
Nous Research has released NousCoder-14B, a new open coding model aimed at the surging demand for AI programming tools.
Google DeepMind Releases GenCeption, a Unified Model
Google DeepMind unveiled GenCeption, a single model that generates scene depth and surface normals at once. What the release actually means.
OpenAI Ships ChatGPT Work, a Universal AI Agent
OpenAI introduced ChatGPT Work, a desktop app merging ChatGPT, Codex and the Atlas browser into one productivity and automation agent.
Gemini CLI Shut Down: Migrating to Antigravity CLI
Gemini CLI is off for consumer accounts: since June 18 Google has moved everyone to the closed-source, Go-based Antigravity CLI, breaking CI/CD pipelines.
GPT-5.6 Work Is Faster and Cheaper Than Claude Cowork
The new GPT-5.6 Work model beats Claude Cowork on report processing: three times faster and four times cheaper in comparative tests.
Rosatom Unveils Kognitron Platform for AI Agents
Russian state corporation Rosatom presented Kognitron, a low-code/no-code platform that lets non-programmers build corporate AI agents.
Anthropic's Claude Now Drives Your Mac by Command
The new Claude Computer Use feature lets the AI control a Mac's mouse and keyboard from text commands, and adds Cowork and Dispatch.
Sber unveils GigaChat 3.5 Ultra for code and text
Sber's new flagship GigaChat 3.5 Ultra is faster, smaller and sharper at writing code, doing math and handling long documents.
OpenAI: ChatGPT Work for your most ambitious tasks
The new ChatGPT Work agent, built on GPT-5.6, can run multi-step tasks on its own, produce deliverables and work through complex projects.
GPT-5.6: OpenAI ships Sol, Terra and Luna
OpenAI launched the GPT-5.6 family: flagship Sol at up to 750 tok/s on Cerebras, Terra at half the price of GPT-5.5, and fast Luna.
Lovable in Talks to Double Valuation to $13.2B
Swedish vibe-coding startup Lovable is reportedly raising ~$300M at a $13.2B valuation, double December's, with Menlo Ventures set to lead.
Grok 4.5: xAI Ships an "Opus-Class" Model at $2/$6
Grok 4.5 is an xAI MoE model trained on Cursor data: "Opus-class" but cheaper and faster, $2/$6 per million tokens, 500K context for coding agents.
GitHub Copilot Desktop App Now Free on Every Plan
GitHub opened its standalone Copilot desktop app to every plan, including Free and Education, on macOS, Windows and Linux, with BYOK and no subscription.
herdr, the "tmux for coding agents", hits #1 on GitHub
herdr, an agent multiplexer for the terminal, hit #1 on GitHub Trending on June 30: a single Rust binary, fully local, 14+ coding agents supported.
X Launches MCP Server for Claude, Cursor and Grok
On June 30 X opened a hosted MCP server: 200+ API endpoints are now available to Claude, Cursor and Grok under your own account's permissions.
Claude Sonnet 5: near-Opus agents at $2/$10
Claude Sonnet 5 is Anthropic's most agentic Sonnet, with a 1M-token window at $2/$10 per million; it is now the default on Free, Pro and in Claude Code.
Agentjacking: A Fake Sentry Bug Hijacks Claude Code
Agentjacking: a forged Sentry error makes an AI agent run attacker code on a developer's machine. 2,388 organizations exposed, 85% success rate.
Mastra npm attack: 140+ packages backdoored
On June 17 the @mastra npm scope was republished with 140+ packages carrying easy-day-js malware that stole crypto wallets. Microsoft blames North Korea.
SpaceX buys Cursor maker Anysphere for $60B
SpaceX is buying Anysphere, the company behind AI code editor Cursor, for $60B in stock — the largest startup acquisition on record.
Ponytail: Claude Code skill rockets to 44K stars
Ponytail is a rules skill for Claude Code, Codex and Cursor that makes AI write 54% less code. It pulled ~44,000 stars in nine days after its 12 June launch.
Claude Fable 5: Anthropic's most powerful model
Claude Fable 5 is Anthropic's most powerful public model: SOTA on coding and agents, $10/$50 per million tokens. Pulled over US export rules, back on 1 July.
Cloudflare acquires VoidZero, the Vite team
Cloudflare has acquired VoidZero, the company behind Vite, Vitest, Rolldown and Oxc. Evan You and his team join Cloudflare; the tools stay open source.
Anthropic files for IPO at $965B valuation
Anthropic filed confidentially for an IPO at a $965B valuation — a week after its $65B Series H, beating OpenAI to the public markets.
Want the long reads instead? Knowledge base in English →