The MCP 2026-07-28 release candidate, locked May 21, drops session management from the protocol core — meaning any MCP request can now hit any server instance behind a standard round-robin load balancer. Simultaneously, Anthropic donated MCP to the new Agentic AI Foundation alongside OpenAI and Block as founding members, with Google, Microsoft, AWS, and Cloudflare as platinum supporters.
Cursor 3.5 (May 20) brings Automations into the Agents Window and adds multi-repo support — finally letting a single automation agent reason across multiple codebases. Five no-repo templates also ship, signalling that Cursor is expanding from coding IDE to general business automation platform.
Windsurf invented the agentic IDE — an editor where the AI reads files, writes code, runs terminal commands, and orchestrates multi-step workflows without leaving your development environment. It’s a genuine innovation. But as Cascade grows more capable and Windsurf absorbs Devin’s autonomous DNA, a structural tension is becoming impossible to ignore: the more agentic an IDE becomes, the more it strains against the architecture that makes it an IDE.
xAI shipped Grok 4.3 on May 4 with 40% lower prices, 1M token context, and 16-parallel-agent orchestration — then followed with Grok Skills on May 18, persistent cross-session expertise that replaces the system-prompt preamble you paste at the start of every chat. The strategy is clear: stop racing GPT-5.5 and Claude Opus 4.7 on raw benchmarks, and build the best cost-per-useful-task ratio on the market.
AI agents now author 16.4% of all test-adding commits in real-world codebases. They’re faster, often broader, and sometimes excellent. They’re also structurally over-mocked, happy-path biased, and capable of producing impressive coverage while testing almost nothing. Here’s what the 2026 research actually shows — and how to use AI tests without fooling yourself.
Most developers using AI to generate code are leaving significant quality on the table. These seven prompting patterns — grounded in how frontier models actually process context — close the gap between ‘it kind of works’ and ‘it ships to production.’
Google launched Antigravity 2.0 at I/O on May 19 — a full rebuild with a standalone desktop app, a Go-based CLI, a public SDK, and Managed Agents API that spins up isolated Linux environments with one call. It scores 76.2% on SWE-bench Verified. It is Google’s most credible attempt yet at terminal-native agent infrastructure, and it still has three structural problems.
Claude Code v2.1.145-146 ships /code-review (replacing /simplify) with optional effort levels, claude agents –json for scripting live session lists, and agent_id/parent_agent_id on OpenTelemetry spans — the missing link for tracing multi-agent executions end to end.
Claude Code v2.1.139 ships two features that change how multi-agent work actually looks: Agent View — a unified dashboard showing every running, blocked, and completed session — and the /goal command, which keeps Claude working autonomously across turns until a defined completion condition holds.
Google I/O 2026 delivered the developer tools story it promised: Firebase Studio launched as a full-stack agent-native development platform, Jules exited beta with free-tier access, and Gemini Code Assist hit general availability. Google’s agentic coding stack is now a real product, not a roadmap.