AI Dev News Digest: January 23rd, 2026

Cursor shipped subagents. The IDE can now spin up parallel specialized agents that each get their own context window, meaning complex refactors don't have to happen sequentially anymore. Meanwhile, Apple is planning to turn Siri into an actual chatbot (codenamed Campos) powered by Google's Gemini, and Meta's Superintelligence Labs delivered its first models internally. The AI coding tools race keeps intensifying.

On the infrastructure side, Azure Functions MCP support hit general availability, Inferact raised $150M to commercialize vLLM, and Neurophos grabbed $110M for light-based AI chips. A Science journal study found 29% of new code in the US is now AI-assisted. Junior dev employment dropped 20% between 2022 and 2025 according to Stanford. The tools are getting better, but the job market implications are getting harder to ignore.

AI Coding Tools

Cursor 2.4 ships subagents. The big feature: parallel specialized agents that each get their own context window. Instead of one agent doing everything sequentially, Cursor can now break down complex tasks and have multiple agents research, run commands, and make edits at the same time. Also adds image generation via Nano Banana Pro. (EveryDev.ai)
Claude Code VS Code extension now GA. Over 3 million installs already. You can watch it work in real-time, review diffs inline, and choose auto-accept or manual approval. Works with Pro, Max, Team, Enterprise, or pay-as-you-go. Requires VS Code 1.98.0+. (EveryDev.ai)
Devin Review launches free AI code review. Cognition Labs shipped a code review tool that transforms large GitHub PRs into organized diffs with bug detection. It groups related changes logically, detects copy/move operations, and flags potential bugs with severity levels. Free for all GitHub repos. (EveryDev.ai)
GitHub CLI gets native Copilot integration. No separate extension needed anymore. Just run gh copilot and it auto-installs. (GitHub Blog)
GitHub Copilot CLI adds Plan mode. New agentic planning capabilities for command-line workflows. (GitHub Blog)

Foundation Models & Labs

Anthropic publishes 2026 Agentic Coding Trends Report. Key finding: developers use AI in about 60% of their work but can only fully delegate 0-20% of tasks. Case studies include Rakuten (99.9% accuracy on 12.5M-line codebase in 7 hours), TELUS (30% faster shipping, 500K hours saved), and Zapier (89% AI adoption, 800+ agents deployed). (Anthropic)
Meta Superintelligence Labs delivers first models. CTO Andrew Bosworth announced at Davos that the lab's early models are "very good" despite being less than six months into development. Media reports suggest codenames Avocado (text) and Mango (image/video). Bosworth called 2025 a "tremendously chaotic year" but said the infrastructure investment is paying off. (Reuters)
Apple planning Siri chatbot for iOS 27. Codenamed Campos, it will replace the current Siri interface and run on a custom Google Gemini model. Apple is paying Google roughly $1B annually and may host the chatbot directly on Google's TPU servers. Expected in late 2026 with iOS 27, iPadOS 27, and macOS 27. (Bloomberg)
OpenAI confirms first consumer device for H2 2026. AI-powered earbuds designed by Jony Ive's io Products team, codenamed "Sweetpea." Open-ear design with 2nm processor for local AI processing. Target: 40-50M units in the first year. (The Information)
Sam Altman teases Codex launches starting next week. OpenAI has "a lot of exciting launches" for Codex coming over the next month. Also announced they're approaching the Cybersecurity High level on their preparedness framework. Initial approach: product restrictions to block coding models from being used for cybercrime. Long-term plan: "defensive acceleration" to help people patch bugs quickly. (Sam Altman)

Developer Platforms & Tools

LM Arena launches Video Arena. Compare AI video models side-by-side via Discord. Supports Sora, Hailuo, Veo 3.1, and others. You vote before seeing which model made which output. Works for text-to-video and image-to-video. (EveryDev.ai)
Vercel teasing something big with v0. Waitlist is live for what looks like a major update. The current v0 already supports GPT-5 and has seen 3.5M+ users to date. (EveryDev.ai)
Azure Functions MCP support hits GA. Built-in OBO authentication, streamable HTTP transport. Supports .NET, Java, JavaScript, Python, TypeScript. Visual Studio 2026 will have Azure MCP Server tools built-in. (Microsoft)
Google Gemini adds free SAT practice tests. Partnered with Princeton Review for real test prep content, not AI-generated questions. Take full-length tests, get instant feedback, and ask Gemini to explain answers. Builds personalized study plans based on weak areas. (EveryDev.ai)

AI Infrastructure Funding

Inferact raises $150M seed at $800M valuation. Commercializing vLLM for AI inference. a16z and Lightspeed co-led. (TechCrunch)
Neurophos raises $110M Series A. Light-based AI chips. Gates Frontier led with M12, Aramco Ventures, and Bosch Ventures. (TechCrunch)
Railway raises $100M Series B. AI cloud platform, TQ Ventures led. (TechCrunch)
Superstate raises $82.5M Series B. Tokenizing securities on blockchain, Bain Capital Crypto and Distributed Global. (Tech Startups)
Cambio raises $18M Series A. AI for commercial real estate, Maverick Ventures led with angels from OpenAI, Anthropic, and Notion. (TechCrunch)

Other Major Funding

Humans& raises $480M seed at $4.48B valuation. Founded by Anthropic, xAI, and Google alums, focused on human-centric AI research. Largest seed round ever. (TechCrunch)
Upscale AI raises $200M Series A. AI networking systems, Tiger Global-led. (Tech Startups)
Zipline raises $600M at $7.6B valuation. Drone delivery expansion. Fidelity, Baillie Gifford, Valor Equity, Tiger Global. (Reuters)
Deepgram raises $130M Series C. Real-time voice AI infrastructure. (TechCrunch)
OpenEvidence raises $250M Series D. AI clinical decision support. (TechCrunch)
Moonshot AI hits $4.8B valuation. Alibaba-backed, up from $4.3B in December. (Bloomberg)

Industry Research

Science journal: 29% of US code now AI-assisted. Up from 5% in 2022. China at 12%, Germany 23%, France 24%, India 20%. AI usage highest among less experienced programmers, but productivity gains (3.6%) accrue mainly to experienced developers. Economic impact: $23-38B annually in the US alone. (Science)
Stanford study: Junior dev employment down 20%. Employment among software developers aged 22-25 fell nearly 20% between 2022 and 2025. Experienced developers also reporting they're losing coding instincts from heavy AI tool reliance. (Stanford)

Platform Updates

Google announces major Gemini education updates at BETT 2026. New features for Google Classroom including lesson planning assistance, content summarization, and contextual explanations. Also new security features like ransomware detection. (Google Blog)
lastminute.com launches MCP Server for Flights. First travel company to ship an MCP server, part of a multi-year AI infrastructure roadmap. (lastminute.com)

Weekend Reading

Apple's Siri Gets the ChatGPT Treatment Apple is planning to transform Siri into a full-fledged AI chatbot called "Campos" later this year, marking a major strategic shift for the company. The chatbot will replace Siri's current interface while keeping the same activation methods (saying "Siri" or holding the side button). It's expected to be unveiled at WWDC in June and ship with iOS 27, iPadOS 27, and macOS 27 in September. Google is reportedly supplying the underlying technology. This is Apple finally admitting that chatbots aren't just a fad—they're table stakes now. The new Siri will do what ChatGPT and Gemini do (web search, content creation, image generation, file analysis) but with Apple's trump card: deep OS integration. It can see your screen, control device settings, and work inside core apps like Mail, Photos, and Xcode. The timing isn't coincidental—OpenAI has been poaching Apple engineers and is building what looks like an AI operating system, which puts Apple's core business in the crosshairs. (Bloomberg)