TechCapsules

LangChain, an agent‑management startup, secured $300 million in a Series C round, elevating its valuation to $3 billion. The funding, led by CapitalG, also saw participation from GIC, DFJ Growth, IVP, and Sapphire Ventures. Founded in 2022 from an open‑source project, LangChain now focuses on AI agent orchestration through its LangSmith platform, which claims 7,000 customers, including Nvidia, Toyota, and Salesforce. The round more than doubles the company’s previous $1.25 billion valuation.

John Langford, a 14‑year Microsoft Research veteran and co‑founder of its AI Frontiers lab, resigned after his team of roughly 50 researchers was absorbed into Microsoft AI, a division led by Mustafa Suleyman that prioritizes scaling over research. Langford claims the new structure ignores cost‑cutting innovations such as his Full Bandwidth Transformer, which could halve AI training and inference expenses. He notes similar breakthroughs have already been validated at ByteDance, yet Microsoft AI remains focused on larger [...]

Six AGIBOT G2 humanoid robots operated continuously for 24 hours in a Shanghai convenience store, completing 1,730 tasks such as customer assistance, coffee preparation, shelf restocking, and cleaning. The trial, livestreamed to over two million viewers, tested the robots’ embodied AI across four daily shifts, from lunch rush to overnight service. Four of the six robots passed the final performance assessment, demonstrating the system’s ability to handle real‑world variability and unexpected obstacles. The event marked the culmination [...]

Google updated Search with Gemini 4 Hydrogen, a new AI model that enhances visual responses and intent understanding. The model delivers image‑rich answers, generates tables for data‑heavy queries, and better dissects complex questions. It also improves back‑linking, surfacing sources users have not visited before, and places source links prominently in AI Mode. The update follows recent changes that prioritize confidence and source visibility, aiming to make everyday searches more intuitive and reliable.

A Game Developers Conference report finds that 40 % of active game developers now use generative AI, up four percentage points from last year. The study surveyed over 2,300 industry professionals, noting that 57 % of users employ AI for text generation, research, prototyping, coding, and presentations. Triple‑A developers report higher pressure to adopt the technology (27 %) than indie developers (9 %). Usage is higher among those 45 + (54 %) and men (45 %) than women (33 %). Despite growing adoption, 55 % of

An AI‑assisted emulator revived an unreleased iOS 4 multitasking interface that Apple abandoned after the iPhone 4 prototype. The interface, resembling Expose, displayed open apps as on‑screen cards and allowed double‑tap Home switching. The video, posted by developer Steve Troughton‑Smith, runs build 8A218 on a QEMU fork, sourced from a prototype root filesystem. Apple’s own AI tools, such as Xcode 27 agents, underscore the growing role of AI in software preservation and development.

ElevenLabs launched an experimental app, ElevenAI, during its 11/11 Summit in New York, offering integrated meeting note‑taking, dictation, and agent‑driven follow‑up actions. The 90‑language transcription tool records meetings, notifies participants, and lets users query past discussions. It also supports industry‑specific terminology and can generate draft emails from notes. The preview app requires early‑access signup, positioning ElevenAI as a team‑focused solution rather than a solo productivity tool. The company simultaneously released its v3 Scribe models for live transcription [...]

Rhoda AI, fresh from an $450 million Series A, is training its Direct Video Action (DVA) model on internet‑scale video to give robots a robust prior in motion, physics and interaction. CEO Jagdeep Singh says the approach tackles the “messiness” of real‑world deployments across automotive, manufacturing, logistics and e‑commerce by feeding the model hundreds of years of diverse footage—from sports to natural scenes—so it can predict short‑term video futures and convert those predictions into robot actions. After pretraining, customers [...]