AI Pulse ๐จ
A bite-sized curation of this week's most important AI news.
โก GitHub launches Vibe Coding tool
โGitHub has launched GitHub Spark in public preview for Copilot Pro+ subscribers. Spark allows users to build and deploy full-stack apps directly from natural language, with zero setup. It includes built-in support for data handling, LLM inference, deployments, hosting, and GitHub auth. Users can integrate models from OpenAI, Meta, DeepSeek, and xAI, then ship apps with a single click.
๐ง Cursor launches Bugbot for better code reviews
โCursor has released Bugbot, an AI tool that improves code quality by catching bugs and vulnerabilities before they ship. Early results show 50%+ resolution rates and up to 40% faster reviews. Engineers say it excels at spotting subtle flaws in AI-generated code, making it a strong fit for modern codebases.
๐ค Claude Code adds sub agents
โAnthropic has added sub agents to Claude Code, enabling developers to spin up specialized AI helpers for focused tasks. Each sub agent can run with its own system prompt, context window, and tools. They can activate automatically when relevant or be manually triggered when needed.
๐ Replit launches Queue for parallel agent work
โReplit introduced Queue, a feature that lets users submit tasks while the Agent is still working. This eliminates idle time and improves workflow efficiency, especially for multi-step or batch processes.
๐๏ธ White House outlines AI plan focused on deregulation
โThe Trump administration released its AI Action Plan, a 23-page proposal focused on accelerating AI through deregulation. It includes loosening environmental rules, restricting funds to states with strict AI laws, and requiring AI models to avoid โtop-down ideological biasโ in order to qualify for federal contracts.
๐ Lovable becomes fastest-growing software startup ever
โSwedish startup Lovable hit $100M ARR in just eight months, setting a new record. With 2.3M active users and 10M projects created, the vibe coding platform also raised $200M at a $1.8B valuation, making it Europeโs newest unicorn.
๐ Gemini 2.5 Flash-Lite now generally available
โGoogle has released the stable version of Gemini 2.5 Flash-Lite, its fastest and cheapest model in the Gemini 2.5 family. It is already in use for tasks like satellite diagnostics and video translation in over 180 languages, with latency reduced by 45%.
๐จ Google unveils Opal for no-code AI mini-apps
โGoogle debuted Opal, an experimental builder that lets users create AI-powered mini-apps using natural language and drag-and-drop editing. Currently in US-only public beta, it targets fast prototyping, productivity apps, and low-code AI tools.
๐๏ธ Meta demos EMG-powered gesture wristband
โMeta has revealed a prototype wristband that lets users control devices with hand gestures. The device reads forearm muscle signals using EMG, predicting movement before it happens. Unlike brain implants, this requires no surgery and works instantly with AI-trained models.
๐ฌ๐ง OpenAI partners with UK on AI adoption
โOpenAI and the UK government have signed an MOU to promote AI in both the public and private sector. OpenAI will expand its presence beyond its current 100-person London office, working with government teams to boost national AI readiness.
๐ DeepMind hits gold medal level at Math Olympiad
โDeepMindโs Gemini Deep Think system has achieved gold medal-level performance at the International Mathematical Olympiad, solving five of six problems.
๐ OpenAIโs reasoning model matches Olympiad gold
โOpenAI announced that its experimental model also reached gold medal-level performance at the IMO. The model solves high-school level math problems across algebra, geometry, and number theory, marking a key step in multi-step reasoning capabilities.