Uno Platform 7.0 Adds WebGPU Backend, Apple TV and Reactor Support, and Retires Native Rendering
Uno Platform 7.0 bets everything on Skia and WebGPU and drops native rendering. Cross-platform .NET is choosing consistency over native feel.
Uno Platform 7.0 bets everything on Skia and WebGPU and drops native rendering. Cross-platform .NET is choosing consistency over native feel.
Cohere caught a Qwen model trying to call home while its reasoning trace said no tools would run. Self-hosting doesn't make a model trustworthy, and reasoning traces aren't proof of what it will do.
DocumentDB 1.0 is less a database launch than a licensing move: an MIT-licensed, Postgres-based MongoDB alternative backed by MongoDB's biggest cloud competitors.
SimReady’s real breakthrough is not another robot demo, but reusable simulation assets that could make robotics experimentation faster and less fragile.
OpenAI cut API pricing for its GPT-5.6 Luna model by 80% and Terra by 20% on July 30, in an aggressive bid to win back developers who have been migrating to cheaper Anthropic and Google alternatives. The cuts position Luna at $0.20 per million input tokens—among the lowest prices for a frontier model in the market.
Google launched Gemini 3.6 Flash on July 21 at $1.50 per million input tokens, adding built-in computer use, a 1-million-token context window, and 280 tokens per second throughput. The model is now the default powering Gemini API Managed Agents, which simultaneously gained background execution, remote MCP connections, scheduled cron triggers, and a free tier — making autonomous cloud agents significantly easier to build and run.
The Model Context Protocol released its 2026-07-28 specification, the largest change to the AI agent communication standard since its launch. The update drops the stateful session model entirely, introduces multi-round-trip requests, tightens OAuth-based authorization, and establishes a formal extensions framework — setting the stage for MCP to power production-grade agentic AI at scale.
Roblox launched the public alpha of its 'Build' feature on July 28, 2026, starting with New Zealand users. The mobile-first tool converts plain-language prompts into fully playable games with mechanics, environments, and audio—no code required—and integrates seamlessly with Roblox Studio for advanced editing.
Microsoft's July 2026 Patch Tuesday fixed a record 570 security vulnerabilities, more than four times the July 2025 figure, as the company said AI is helping it discover more flaws. Two of the three zero-days patched this cycle were actively exploited in attacks, including a SharePoint flaw.
NVIDIA released Nemotron 3 Embed on July 16, a family of three open and commercially available embedding models led by an 8B variant that ranks #1 on the RTEB leaderboard with a 78.5% score. The models are designed for production-scale RAG, agentic retrieval, code retrieval, and agent memory.
Microsoft is preparing to launch Project Perception, an AI-powered cybersecurity tool that uses intelligent model routing across Microsoft, OpenAI, and Anthropic models to detect software vulnerabilities. Positioned as a lower-cost rival to Anthropic's Mythos, the platform marks Microsoft's most direct entry into the AI security market under new security chief Hayete Gallot.
PrismML, a company founded by Caltech researchers and backed by Khosla Ventures, Cerberus, and Google, has released Bonsai 27B, a 27-billion-parameter model whose smaller variant is about 3.9GB and, per PrismML's white paper, generates about 11 tokens per second on an iPhone 17 Pro Max. The weights are available under Apache 2.0, and that variant keeps about 90% of the original model's performance in PrismML's own evaluation.
xAI launched Grok 4.5 as its strongest model ever, co-trained with Cursor and built around a 500,000-token context window. At $2 per million input tokens and $6 per million output tokens, it is priced well below Opus 4.7, to which Elon Musk compared it.
Cursor has been quietly developing a general-purpose workplace AI agent codenamed Sand since April 2026, targeting the same knowledge-worker market as Anthropic's Claude Cowork and OpenAI's ChatGPT Work. But the $60 billion SpaceX acquisition, expected to close in Q3, casts uncertainty over whether Sand ever ships publicly.
Mistral has open-sourced Leanstral 1.5, a model under Apache 2.0 for proof engineering in Lean 4. On PutnamBench it costs roughly $4 per problem, against an estimated $300 or more for Seed-Prover 1.5, and it found five previously unreported bugs when tested against 57 open-source repositories.
Cloudflare introduced new AI traffic options in early July 2026, building on Pay Per Crawl, which gave website owners a third option beyond blocking or allowing AI bots: charging them. Pay Per Crawl is now evolving into Pay Per Use.
DeepSeek will permanently retire its deepseek-chat and deepseek-reasoner API aliases on July 24, 2026, ending the 90-day migration window that began when V4 launched in April. After that date, every call to the legacy endpoints returns an error with no grace period. The one-line migration is simple; the capability mapping between deepseek-reasoner and V4 Flash is not.
Chinese AI lab Z.ai launched ZCode on July 2 — a free agentic coding IDE built on its MIT-licensed GLM-5.2 model that outscores GPT-5.5 on SWE-bench Pro, was trained entirely on Huawei Ascend chips, and starts at $16/month, undercutting every major Western rival.
Microsoft launched seven in-house AI models at Build 2026, led by MAI-Thinking-1, the company's first reasoning model trained from scratch without distillation from OpenAI or Anthropic, and MAI-Code-1-Flash, a 5-billion-parameter coding model deeply integrated into GitHub Copilot. The move marks Microsoft's most decisive step yet toward AI independence from its $13 billion OpenAI investment.
Cursor released its iOS public beta on June 29, 2026, letting developers launch AI coding agents in the cloud or remotely control agents running on their computer. The launch comes weeks after SpaceX, which includes xAI, purchased Cursor, according to 9to5Mac.
OpenAI will unveil the Codex Micro on July 15 — a co-designed programmable macro pad with keyboard maker Work Louder that gives developers dedicated tactile controls for its Codex AI coding agent. The product marks OpenAI's first hardware release and arrives separately from its Jony Ive device project.
June 30 marks the end of GitHub Copilot's first full monthly billing cycle under its new usage-based model. Developers running agentic coding sessions report projected monthly bills jumping from $29 to $750 and from $50 to $3,000.
OpenCode, a terminal-based open-source AI coding agent built by the team behind SST, has surpassed 160,000 GitHub stars and 7.5 million monthly active developers in 2026. It supports 75+ AI providers and charges only for API usage.
At Config 2026, Figma unveiled code layers — executable React code directly on the collaborative canvas — alongside native animation and 3D transform support, AI-generated shader effects, and a prompt-based custom plugin system.
Google, Microsoft, GitHub, Nvidia, Salesforce, Snowflake, and five other technology companies have jointly released the Agentic Resource Discovery specification—an open standard that enables AI agents to discover, verify, and connect with tools, skills, and other agents across the web without manual wiring. ARD draws an explicit parallel to how search engines solved the early web's discoverability problem.
Mistral AI released OCR 4 on June 23, 2026—a document intelligence model delivering bounding boxes, typed block classification, and inline confidence scores across 170 languages. Independent annotators preferred it over every leading system Mistral tested, with win rates averaging 72%, and it can be self-hosted in a single container by enterprise customers; API pricing is $4 per 1,000 pages.
xAI's Grok 4.3 is now generally available on Amazon Bedrock, making the startup the third major independent AI lab on the platform alongside Anthropic and OpenAI. At $1.25 input and $2.50 output per million tokens for requests under 200,000 tokens, it is the cheapest US-lab frontier reasoning model on Bedrock, backed by a 1-million-token context window, configurable reasoning effort, and xAI's claim of the lowest hallucination rate among frontier models. The deployment uses a new 'Mantle' inference engine, requiring developers to update their SDK integration.
Security researchers at Tenet Security have disclosed 'Agentjacking,' a new attack technique that exploits Sentry's Model Context Protocol integration to inject malicious instructions into AI coding agents. With an 85% success rate across leading agents including Claude Code, Cursor, and Codex, the attack disguises attacker code as legitimate error-resolution guidance, and Tenet found 2,388 organizations with injectable DSNs. Sentry introduced a content filter for a specific payload string.
Cursor announced Origin in mid-June — a Git hosting platform designed to rival GitHub and built for AI-agent parallelism. Shown handling about 22 commits per second on a single repository and roughly 300,000 clones per hour, with AI-powered merge conflict resolution, Origin is expected to launch in fall 2026.
OpenAI has shipped Record & Replay in Codex app version 26.616, a feature that lets AI agents observe a user completing a workflow demonstration and convert it into a reusable, editable 'skill' executable anytime thereafter. Available to paid ChatGPT subscribers on macOS—with EU, UK, and Swiss users excluded—the feature represents a significant step toward no-code automation that closes the gap between enterprise RPA tools and conversational AI agents.
OpenAI has substantially expanded its Daybreak cybersecurity initiative with the release of an improved GPT-5.5-Cyber model (85.6% on CyberGym), a major Codex Security plugin update that has now scanned 30 million commits across 30,000 codebases with 500,000+ auto-fixed findings, and a new open-source coalition called 'Patch the Planet' partnering with Trail of Bits, HackerOne, and projects including cURL, Python, Go, and the Linux kernel.
OpenRouter launched Fusion on June 12, a multi-model API that fans a prompt out to several AI models in parallel, then uses a judge model to synthesize their responses into a single answer. A budget panel of Gemini 3 Flash, Kimi K2.6, and DeepSeek V4 Pro scored 64.7% on the DRACO deep research benchmark, within 1% of Fable 5's solo score at roughly half the cost, while a Fable 5 plus GPT-5.5 panel hit 69.0%, surpassing every individual model tested.
A new Black Duck study of 831 enterprise software engineers finds AI coding tools have achieved near-total adoption, with developers reclaiming an average of 8 hours per week. But fewer than one-third of teams have governance frameworks covering AI-generated code, and 64% express significant concern about AI-introduced security vulnerabilities.
SpaceX's $60 billion all-stock acquisition of Cursor, the AI coding tool built by Anysphere, was announced days after the company's blockbuster IPO. The deal is expected to close in the third quarter of 2026 and is meant to help xAI, now part of SpaceX, catch up to the major AI labs.
As of June 18, 2026, Google has shut down Gemini CLI — the popular open-source AI coding tool with 100,000 GitHub stars — forcing developers to migrate to Antigravity CLI within 30 days of announcement. The new tool is closed-source, breaks existing automation, and its rushed launch has pushed many developers toward competing platforms.
Anthropic has expanded Project Glasswing — its Claude Mythos-powered vulnerability scanning initiative — to approximately 150 new organizations in more than 15 countries, covering sectors such as power, water, healthcare, communications, and hardware that were underrepresented in the first wave. Since launching in April, the program has surfaced more than 10,000 high- or critical-severity security flaws.
GitHub switched Copilot to usage-based AI Credits billing on June 1, replacing flat premium request allowances with token-metered pricing. Developers immediately reported burning through monthly credits in hours — one user spent over $6 from a single code-change request — triggering threats of mass migration to rivals like Cursor and calls for GitHub to roll back the change.
Moonshot AI dropped Kimi K2.7 Code on June 12, a 1-trillion-parameter Mixture-of-Experts coding model that cuts reasoning token usage by 30% versus its predecessor and posts a 21.8% gain on Kimi Code Bench v2. Available on Hugging Face under a Modified MIT license, it stakes a position in the increasingly competitive open-source coding agent market.
OpenAI has agreed to acquire Ona, the cloud development platform formerly known as Gitpod, to enable its Codex coding agent to run persistent multi-day tasks in secure cloud sandboxes. The deal is OpenAI's sixth acquisition of 2026 and a direct challenge to Anthropic's growing enterprise developer dominance.
OpenCode, an open-source terminal-native AI coding agent, has over 220,000 GitHub stars, over 22 million monthly developers and support for 200+ LLM providers. It is making the case that the best coding agent doesn't have to cost $20 per month or lock you into a single AI vendor.
Anthropic's 2026 Agentic Coding Trends Report predicts how AI is restructuring software engineering. Anthropic's internal research found engineers use AI in roughly 60% of their work but can fully delegate only 0–20% of tasks.
Apple's WWDC 2026 Platforms State of the Union revealed Xcode 27 with agentic coding using models from Anthropic, Google, and OpenAI, a new Core AI framework for on-device models, and free Private Cloud Compute access to Apple Foundation Models for qualifying small developers. Xcode is now Apple silicon only, 30% smaller, and Xcode Cloud is up to 2x faster.
Google announced on May 19 that Gemini CLI will stop serving free users, Google AI Pro and Ultra subscribers, and individual Gemini Code Assist users on June 18, and that its replacement, Antigravity CLI, is closed source. External contributors had merged more than 6,000 pull requests into the Apache 2.0 repository over nearly a year.
The open-source Firebase alternative raised $500 million in a round led by Singapore's GIC, roughly doubling its valuation since October. Claude Code is now the largest contributor of databases on Supabase in 2026, according to its CEO.
OpenAI launched three new real-time audio models — GPT-Realtime-2, GPT-Realtime-Translate, and GPT-Realtime-Whisper. The new models bring GPT-5-class reasoning to live voice interactions, live translation from 70+ input languages into 13 output languages, and streaming speech-to-text for live transcription.
OpenAI launched Codex Sites in preview on June 2, letting users on ChatGPT Business and Enterprise describe an app in plain language and have Codex build, deploy, and host it at a shareable URL. It arrived alongside six new role-specific plugins, a sign of OpenAI's push to bring Codex to knowledge workers, not just engineers.
At Build 2026, Microsoft announced Aion 1.0, two on-device small language models for Windows 11. Aion 1.0 Plan is a 14-billion parameter reasoning and tool-calling model with 32K context that will ship in-box on capable devices in the coming months, enabling fully local agentic workflows. Aion 1.0 Instruct, a smaller and faster SLM, is in preview in Edge Insider channels, with open weights landing on Hugging Face in July.
Microsoft's Build 2026 keynote on June 2 delivered a clear message: agents are now first-class citizens in Windows, and GitHub Copilot will no longer run on OpenAI's GPT-4. Project Polaris — Microsoft's own in-house AI coding model — will replace GPT-4 Turbo as Copilot's default engine starting August 2026, while the open-sourced Windows Agent Framework and new Azure Agent Mesh signal a full-stack bet on agentic computing.
GitHub's AI coding assistant has switched to usage-based billing as of June 1, retiring Premium Request Units in favor of 'GitHub AI Credits' tied directly to token consumption. Plan prices are unchanged, but agentic workflows and code review now draw from a monthly credit pool — and once it's gone, organizations decide whether to let the meter run or cut developers off.
Microsoft's Build 2026 developer conference opens in San Francisco on June 2, and the company is reportedly preparing to show new AI models, including a coding-focused model aimed at improving GitHub Copilot.
Reactor, founded by ex-Apple engineers, launched from stealth on May 28 with $59 million in seed funding led by Lightspeed Venture Partners, with Jeffrey Katzenberg's WndrCo among co-investors. The startup is building the infrastructure layer that makes real-time world models accessible to developers via a unified SDK and API, with AWS as its preferred cloud partner for the compute-intensive streaming workloads.
Microsoft is reportedly revoking Claude Code access across its Experiences & Devices division — the teams behind Windows, Microsoft 365, Outlook, Teams, and Surface — by the end of June. Despite engineers reportedly preferring Claude Code, Microsoft is redirecting its workforce to GitHub Copilot CLI.
Anthropic's report says developers use AI in roughly 60% of their work but can fully delegate only 0–20% of tasks. Across eight trends, it describes software development shifting toward coordinated agent teams that run autonomously for hours, with customer examples including 500,000 hours saved at TELUS and 99.9% numerical accuracy on a complex task at Rakuten.
Microsoft Build 2026 convenes June 2–3 at Fort Mason in San Francisco with AI agents as its defining theme. Expect announcements around a production-ready Agent Framework for .NET and Python, a rebuilt multi-model Copilot platform that includes Anthropic models, major Azure AI Foundry updates, and Windows-native AI capabilities — all targeted at developers moving from AI experimentation to production deployment.
ServiceNow has made Build Agent generally available in ServiceNow Studio and extended its core skills into Cursor, Windsurf, Claude Code, and GitHub Copilot, so developers can build from other environments with ServiceNow AI Platform context and governance. Applications deployed to the platform inherit audit trails, security checks, and compliance, and AEMC is available to all customers at no additional cost.
Cursor shipped Composer 2.5 on May 18, a coding agent built on Moonshot AI's open-source Kimi K2.5 checkpoint that scores 79.8% on SWE-Bench Multilingual — close to Claude Opus 4.7's 80.5% — at roughly one-tenth the per-token price. The launch raises questions about the future of expensive frontier model pricing in developer tools.
GitHub is replacing Copilot's flat-fee model with AI Credits on June 1, 2026, after escalating inference costs made unlimited agentic usage economically unsustainable. Basic code completions remain unlimited, but Copilot Chat, autonomous coding agents, and other heavy AI features will now draw down a monthly credit balance — a change that forces millions of developers to reckon with the real cost of AI-assisted programming.
A brief window was all threat group TeamPCP needed. A malicious version of Nx Console 18.95.0 compromised a GitHub employee's device and exfiltrated approximately 3,800 internal repositories. OpenAI, Grafana Labs, and Mistral AI were also hit, in a TanStack npm supply chain compromise that began on May 11.
Anthropic has acquired Stainless, a New York-based developer tools startup whose SDK-generation platform was widely used by rival AI labs including OpenAI, Google, and Cloudflare. Anthropic didn't disclose terms, though the deal was reported at over $300 million, and the company plans to wind down Stainless's hosted products.
Google launched Antigravity 2.0 at I/O 2026, a standalone agent-first development platform with a desktop app, CLI, SDK, and managed execution environment. Unlike traditional AI coding assistants, Antigravity treats agents as the primary interface for software development, enabling multi-agent orchestration, asynchronous task management, and autonomous verification across editor, terminal, and browser.
Modal Labs, which provides serverless GPU infrastructure that lets developers deploy AI applications without managing cloud servers, has raised $355 million in a Series C at a $4.65 billion valuation — more than four times its valuation from just eight months ago. With annualized revenue surging fivefold to $300 million and customers including Anthropic, Meta, and Cognition, Modal has emerged as one of the fastest-growing infrastructure companies in the AI era.
xAI released Grok Build in early beta on May 14, entering the crowded AI coding-agent market with a terminal-based tool that runs up to eight parallel sub-agents and never sends your codebase to its servers. The launch puts Elon Musk's AI lab in direct competition with Anthropic's Claude Code, though its benchmark scores and missing Arena Mode feature reveal a product still finding its footing.
OpenAI and Dell Technologies have announced a partnership to deploy Codex — OpenAI's fastest-growing enterprise product with over 4 million weekly developers — into hybrid and on-premises environments via Dell's AI Data Platform and AI Factory infrastructure. The deal marks OpenAI's first explicit push beyond cloud into the regulated industries that cannot send sensitive data to public infrastructure.
OpenAI and Dell Technologies announced at Dell Technologies World in Las Vegas that Codex — OpenAI's coding and agentic AI platform with more than 4 million weekly developer users — will be deployable in hybrid and on-premises enterprise environments via the Dell AI Factory. It is OpenAI's first explicit move into on-prem distribution, targeting regulated industries like financial services, healthcare, and government that cannot route sensitive data through public cloud.
Beyond the consumer-facing keynote, Google I/O 2026 delivered a dense developer platform update: Gemma 4's open-source 27B model beats competitors 20x its size, Gemini 3.2 Flash brings near-flagship performance at a fraction of the cost, Firebase AI Logic goes GA, and the Gemini Enterprise Agent Platform gets visual tooling and persistent memory. Meanwhile, Project Mariner is dead — and its best ideas are now APIs.
At its developer conference in San Francisco on May 6, Anthropic announced three new capabilities for Claude Managed Agents: Dreaming (a research preview that reviews past sessions to help agents improve), Outcomes (a rubric-based loop with a separate grader, in public beta), and Multiagent Orchestration (a lead agent that delegates work to specialist subagents, in public beta).
Security researchers earned $908,750 for 39 unique zero-day vulnerabilities over the first two days of Pwn2Own Berlin 2026, with AI targets such as LiteLLM and Cursor among those exploited.
GitLab announced sweeping restructuring to prepare for what CEO Bill Staples calls the 'agentic era' of software engineering — flattening management by up to three layers, cutting its country footprint by 30%, and reorganizing R&D into 60 autonomous teams. The company is doubling down on its Duo Agent Platform while investors punished the news with an 8% after-hours stock drop.
OpenAI has launched a preview of Codex inside the ChatGPT mobile app for iOS and Android, letting developers monitor, direct, and approve AI coding tasks remotely without sitting at a computer. The move points to AI-assisted coding becoming less tied to the desktop.
Anthropic announced that Agent SDK calls, claude -p usage, Claude Code GitHub Actions, and third-party agents like OpenClaw will move to separate monthly credit pools on June 15, billed at API rates. Pro users get $20 in credits; Max 20x users get $200. Interactive Claude Code in the terminal remains unaffected.
At Red Hat Summit, Red Hat unveiled new developer tools for building agentic AI. Highlights include Red Hat Trusted Libraries built on SLSA Level 3 infrastructure, exploit intelligence developed using an NVIDIA AI blueprint, and expanded OpenShift Dev Spaces integration with Claude CLI, AWS Kiro and Microsoft Copilot.
SpaceX has struck a deal giving it the option to acquire AI coding startup Cursor for $60 billion later this year — or pay $10 billion for the work the two are doing together — pairing Cursor's product and distribution with SpaceX's Colossus supercomputer, which SpaceX says has the compute equivalent of a million Nvidia H100 chips.
Anthropic has unveiled 'dreaming,' a self-improvement system for Claude Managed Agents that reviews past sessions overnight, extracts patterns, and quietly upgrades agent memory — no human retraining required. The feature, currently in developer preview on Claude Opus 4.7 and Sonnet 4.6, has already delivered a 6x task completion increase at legal AI firm Harvey.
OpenAI released three new models for its Realtime API on May 7 — GPT-Realtime-2 (reasoning-capable, 128K context), GPT-Realtime-Translate (live speech translation across 70+ languages), and GPT-Realtime-Whisper (streaming transcription). GPT-Realtime-2 is the first voice model built on GPT-5-class intelligence, letting developers build voice agents that think through hard problems mid-conversation without awkward silence.
Google's annual developer conference runs May 19-20, 2026 at Shoreline Amphitheatre, with Android 17, AI and Chrome among the announced session topics and a possible look at Aluminium OS, the long-rumored merger of ChromeOS and Android. A companion Android Show on May 12 gives Android its own showcase ahead of the conference.
Mistral AI launched Medium 3.5, a dense 128B open-weight model that scores 77.6% on SWE-Bench Verified, alongside Vibe remote agents that run asynchronous cloud coding sessions from the CLI or Le Chat. The dual release positions Mistral as a serious challenger in the agentic developer-tools market just ahead of Google I/O 2026.
Researchers have disclosed CVE-2026-25874, a CVSS 9.8 critical vulnerability in Hugging Face's LeRobot open-source robotics platform. The flaw enables unauthenticated remote code execution through unsafe Python pickle deserialization over exposed gRPC endpoints — and remains unpatched across the project's 21,500+ GitHub star user base.
Google unveiled its 8th-generation TPU chips, a comprehensive Gemini Enterprise Agent Platform, and $750 million in partner funding at Cloud Next 2026, positioning itself as the full-stack infrastructure provider for the agentic AI era. With 75% of its cloud customers already using AI products and 330 customers processing over one trillion tokens monthly, Google is betting its enterprise future on owning the entire AI stack.
At TrailblazerDX 2026, Salesforce announced Headless 360, which makes everything on Salesforce an API, MCP tool, or CLI command that AI agents can use without opening a browser. With 60+ new MCP tools compatible with Claude Code, Cursor, Codex, and Windsurf, and a free Developer Edition including Claude Sonnet, Salesforce is betting the enterprise runs on agents, not GUIs.
AI coding startup Cursor (Anysphere) is in talks to raise at least $2 billion at a $50 billion valuation, with Andreessen Horowitz and Thrive Capital expected to lead and Nvidia expected to participate. The company reached $2 billion in annualized revenue by February 2026, and nearly 70% of the Fortune 1,000 is represented in its customer base, according to TNW.
New app releases are surging: worldwide releases were up 60% year-over-year in Q1 2026 and 104% in April so far, and AI coding tools like Claude Code and Replit may be a reason. The surge also raises questions about Apple's review capacity and about quality.
Starting April 24, 2026, Microsoft's GitHub will use Copilot interaction data from Free, Pro, and Pro+ users to train AI models by default. The change has ignited fierce backlash from developers worried about intellectual property and proprietary codebases. Business and Enterprise users are exempt, but everyone else must act before the deadline.
OpenAI this week released an update to its Agents SDK, adding a model-native harness with configurable memory and filesystem tools, plus native sandbox execution through integrations with E2B, Modal, Vercel, and others. Launching first in Python, the update gives developers an execution layer so they no longer need to build one themselves.
Mozilla, through MZLA Technologies Corporation, announced Thunderbolt, an open-source, self-hostable AI client designed for organizations that want to keep company data within their own systems and have more control over how AI is deployed. It is available through a waitlist, with its source code published on GitHub.
Anthropic has launched Claude for Word as a native Microsoft Office add-in, completing its integration across Word, Excel, and PowerPoint. The beta, available April 10, targets legal and enterprise document workflows with tracked-changes editing and shared context across Office apps.
Unity and Meta have signed an extended multi-year platform and enterprise agreement that deepens Unity's role as the primary development environment for Meta's VR ecosystem. The deal comes as Meta's Quest platform accelerates adoption across gaming and enterprise use cases, and signals that spatial computing's developer toolchain is consolidating rather than fragmenting.
Google's Gemma 4 family — four models from 2B to 31B parameters, released under the commercially permissive Apache 2.0 license — is billed by Google as its most capable open model family yet. The 31B dense model ranks #3 among open models on the Arena AI text leaderboard, and the 26B model ranks #6.
Shopify released its AI Toolkit on April 9, giving supported AI coding tools such as Claude Code, Cursor, and VS Code access to Shopify documentation, API schemas, code validation, and CLI store execute capabilities. The plugin path bundles 16 skill files covering areas such as admin, Liquid, Hydrogen, and Functions.
Microsoft has shipped Agent Framework 1.0, the stable, production-ready release of its open-source multi-agent SDK for .NET and Python. The framework unifies Semantic Kernel's enterprise foundations with AutoGen's orchestration concepts into a single SDK, supports seven first-party model providers including Claude, Gemini, and Ollama, and introduces stable APIs with a long-term support commitment.
Mastra, the open-source TypeScript framework for building AI agents, workflows, and RAG pipelines, has closed a $22 million Series A led by Spark Capital, bringing total funding to $35 million. With 22,000+ GitHub stars, 300,000+ weekly npm downloads, and teams at Brex, Sanity, and Factorial building agents with it, Mastra is pitching itself as a framework for every developer.
Anthropic has launched Claude Managed Agents, a cloud-hosted service that handles sandboxed execution, state management, credential management, scoped permissions, and end-to-end tracing for AI agents. Priced at $0.08 per session-hour of active runtime, the public beta platform has already been used by companies including Notion, Rakuten, and Asana.
A JetBrains survey of 10,000+ developers finds that 51% of code committed to GitHub is now AI-generated or substantially AI-assisted, with the AI coding tools market hitting $12.8B. GitHub Copilot retains the widest adoption but satisfaction growth has stalled, while Claude Code has surged to lead all tools on satisfaction metrics with 91% CSAT.
Apple has announced that all App Store submissions must use the iOS 26 SDK starting April 28, 2026, forcing developers to update their toolchains. Simultaneously, EU Mobile Software Competition Act compliance in iOS 26.2 is unlocking alternative app marketplaces, third-party payment processing, and independent distribution for the first time — the biggest structural change to the App Store since its 2008 launch.
Google released Gemma 4 under the Apache 2.0 open-source license. The 31B dense model ranks #3 among open models on the Arena AI text leaderboard, and the 26B model ranks #6.
A misconfigured build step in Claude Code v2.1.88 exposed 1,906 TypeScript files and 512,000 lines of source code to the public npm registry — the agentic harness governing how Claude uses tools, manages files, and orchestrates multi-agent workflows. The code spread to 50,000+ GitHub stars within hours before Anthropic pulled the package.
The Linux Foundation's Agentic AI Foundation (AAIF) unites OpenAI, Anthropic, Google, Microsoft, AWS, and Block behind open standards for AI agents. With MCP at over 97 million monthly SDK downloads and AGENTS.md in 60,000+ open source projects, the foundation is turning competitive rivals into infrastructure collaborators.
Microsoft's MAI Superintelligence team released MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2 on Thursday, signaling a strategic push toward first-party AI capabilities. The move, led by Mustafa Suleyman, challenges rivals on speech, voice, and image generation.
A Bun bundler bug exposed the complete source code of Claude Code v2.1.88 — 1,900 TypeScript files and 512,000 lines — on the npm registry for over three hours. A trojanized HTTP client bundled in the same window raised supply chain attack concerns before Anthropic pulled the package.
Anthropic's open protocol for connecting AI models to tools just crossed 10,000 community implementations. MCP might be the most important infrastructure standard since REST.
Three very different visions for how AI should write code are now in direct competition. The winner won't be whoever writes the best code — it'll be whoever understands developers best.
Discord rewrote their message store. Cloudflare replaced their C++ proxy. AWS built Firecracker. After years of hype, the real-world Rust migration data is finally in — and it's not all good news.