8 stories · curated from the last 24 hours
At yesterday's Android Show: I/O Edition, Google recast Android as an AI-first platform. The centrepiece is Gemini Intelligence — a system-wide layer that works across apps, browsers, and devices to autonomously handle tasks like booking a spin class or building a shopping cart from a notes list, with user confirmation before acting.
Google also revealed Googlebooks: Gemini-native laptops shipping this fall with OEM partners including Acer, Asus, Dell, HP, and Lenovo. A new Create My Widget feature lets users describe a widget in plain text and have Gemini generate it on the fly. Full Google I/O keynote is set for May 19.
Announced at SAP Sapphire, Anthropic and SAP are embedding Claude as the primary reasoning engine across SAP's AI-enabled solution portfolio. Claude-powered agents will connect to the SAP Business AI Platform to operate in context with SAP data, taking real actions across finance, HR, procurement, and supply chain workflows — targeting SAP's hundreds of thousands of enterprise customers.
Anthropic launched the Claude Platform on AWS, bringing the full Messages API, Files API, Message Batches API, Managed Agents, and code execution to Anthropic-managed infrastructure accessible through AWS billing and IAM — letting teams stay entirely within their existing cloud footprint.
Separately, Anthropic doubled Claude Code's five-hour rate limits across Pro, Max, Team, and seat-based Enterprise plans, and removed the peak-hours cap reduction for Pro and Max. API rate limits for Claude Opus models were also raised significantly. Claude for Excel, PowerPoint, and Word reached general availability; Claude for Outlook entered public beta for all paid plans.
OpenAI's Daybreak Initiative, powered by GPT-5.5-Cyber and Codex Security, is a direct answer to Anthropic's Project Glasswing / Claude Mythos security program. Anthropic's Claude Mythos recently helped Mozilla identify and fix more than 270 Firefox vulnerabilities; OpenAI is now offering its cyber-specialised model to vetted security teams and, separately, to EU access as well.
Both initiatives come as Google's Threat Intelligence Group confirmed the first known case of an attacker using AI to discover and exploit a zero-day vulnerability — a milestone that is accelerating demand for AI-native defensive tooling.
OpenAI announced a new professional services business backed by $4 billion in investment, targeting large enterprise customers who need hands-on help deploying and customising GPT-5.5 and Codex-based solutions. The move mirrors the playbook of major cloud and consulting firms and signals that OpenAI is no longer purely a product company — it is actively competing for implementation budgets alongside systems integrators.
Microsoft added xAI's Grok 4.1 Fast to Copilot Studio as a selectable model provider, available to US-based makers once enabled by organization administrators. The addition extends the multi-model strategy Microsoft has been building — giving enterprise builders access to Grok's real-time X data advantage alongside existing options from OpenAI, Anthropic, and Google inside the same low-code agent builder.
Signadot launched a new skill that lets Claude Code, OpenAI Codex, and Cursor validate code changes against production-like Kubernetes environments before handing work back to developers. The skill connects agents via an MCP server, spins up a Signadot Sandbox containing only the modified service (everything else is shared from the baseline cluster), runs tests, reads results, and iterates — closing the loop on one of the sharpest limits of current coding agents: they write good code but can't verify it in complex distributed systems.
Rather than one agent capturing the market, Cursor, Claude Code, and OpenAI Codex are converging into an orchestration / execution / review stack that teams run together. Cursor shipped a rebuilt parallel-agents UI; OpenAI published an official plugin running inside Claude Code. A Pragmatic Engineer survey of 906 engineers (February 2026) put Claude Code at #1 with a 46 % "most loved" rating; SemiAnalysis estimates it accounts for ~4 % of all public GitHub commits, with projections toward 20 % by year-end.