AI Pulse 🚨
A bite-sized curation of this week's most important AI news.
🏗️ Lovable nears $2B valuation. According to the Financial Times, Swedish AI startup Lovable is close to reaching a $2 billion valuation in its latest funding round. The Stockholm-based company has gained momentum for its “vibe coding” approach and is becoming a standout in the European AI ecosystem.
🎨 Figma files for IPO. According to The Verge, Figma has filed for an IPO under the ticker “FIG” after its $20B Adobe deal collapsed in 2023. Revenue grew to $228.2M from $156.2M year-over-year. CEO Dylan Field plans to “double down” on AI, stating: “AI spend will potentially be a drag on our efficiency for several years, but AI is also core to how design workflows will evolve going forward.”
🧠 Does ChatGPT dull your thinking? A recent MIT study suggests using ChatGPT to write essays weakens brain connectivity. Researchers tracked participants using EEG headsets and found those who used AI retained far less of what they wrote—0% could quote from their essay afterward, compared to 89% in the non-AI group.
🤔 AI might not be the problem. Anne-Laure Le Cunff’s response highlights MIT study limitations (small sample size, artificial task), and offers practical advice: stay mentally active while using AI, write first and refine later, and use tools that challenge rather than replace your thinking.
🧩 LangChain introduces Context Engineering framework. The new LangChain blog post presents a clear framework for managing long-running agents. It breaks down context strategies into four types: write (save context externally), select (retrieve relevant info), compress (remove excess), and isolate (split across agents).
🍿 New AI buzzword! "Context Engineering" gains traction. Philipp Schmid expands on the idea, arguing the future of AI isn’t about perfect prompts—but delivering the right context at the right time. Most failures, he says, come from missing context, not model limitations.
📱 Cursor launches agents on web and mobile. Cursor, the AI-powered coding tool, now supports agents across web, iOS, and Android. Users can build code, run background tasks, and collaborate on the go. Features include mobile access, Slack integration, and team support.
🧠 Anysphere hires Anthropic’s Claude Code leaders. As reported by The Information, Anysphere has recruited key figures from Anthropic’s Claude Code team—another sign of intensifying AI talent competition across companies.
📊 Perplexity launches Max subscription. Perplexity Max includes unlimited Labs usage for building dashboards, spreadsheets, and web apps, plus access to premium models like o3-pro and Claude Opus 4.
⚙️ Cursor vs Claude Code: devs debate best AI coding tool. In a Product Hunt thread, developers weigh Cursor’s fast iteration and web app building tools against Claude Code’s power and depth. One common friction point: Claude’s pricing.
🧑💻 Built with Claude: 20k-line macOS app with <1k lines hand-coded. In a deep dive blog post, a developer walks through how they built Context, a full macOS app for debugging MCP servers, using Claude Code. The post details what Claude did well, where it needed guidance, and why well-written specs are still critical for quality results.
🏥 Microsoft’s diagnostic AI outperforms doctors in complex cases. Microsoft researchers say their AI system—combining OpenAI’s o3 model with a diagnostic orchestrator—correctly diagnosed 8 out of 10 complex medical cases, compared to just 2 of 10 for human doctors. The system is not yet clinically deployed but is being positioned as a more efficient, potentially lower-cost solution to augment healthcare decision-making.
🏛️ Meta forms Superintelligence Labs. In a memo, Zuckerberg announced Meta Superintelligence Labs (MSL), a new org to consolidate its foundation models, FAIR research, and applied AI efforts. MSL is led by a star-studded team including Alexandr Wang and Nat Friedman, and is tasked with building “personal superintelligence for everyone.”
📈 Superhuman acquired by Grammarly. Superhuman is joining forces with Grammarly to build a unified AI productivity suite—starting with email. Grammarly plans to leverage its massive app integration footprint (500,000+ apps) to embed smart agents throughout daily workflows.
🏫 Google brings Gemini to the classroom. At the ISTE edtech conference, Google unveiled 30+ AI features for educators and students. Highlights: Gemini-powered lesson planning, custom AI tutors (“Gems”), reading buddies, and classroom-safe video creation tools. The rollout also includes features for student progress tracking and Chromebook management.
🧑💻 Google issues internal guidance on AI-assisted coding. According to 9to5Google, Google has released official recommendations for developers using AI in production. With 30%+ of its code now AI-generated, the company emphasizes human review, spec-first workflows, and clear team collaboration to ensure quality.
