42 noviniek
The July 19, 2026 Claude Code update changes /verify and /code-review skill behavior — Claude no longer triggers them autonomously, only on direct invocation. Increases agentic pipeline predictability.
Cursor updated CursorBench to v3.1 focused on codebase understanding, bug-finding, and code review across 36 models. Fable 5 Max leads at 72.9%, while Composer 2.5 offers the best value at $0.55 per task.
Cursor released its native iOS app for iPhone and iPad into public beta for all paid plans, letting users spawn and manage cloud agents, remote-control a desktop session, and code by voice from mobile.
Anthropic raised Claude API rate limits across the board — Sonnet and Haiku now match Opus at every tier — and collapsed usage tiers into three (Start, Build, Scale), powered by a new SpaceX compute deal.
Vercel's AI Gateway adds beta support for realtime voice agents, text-to-speech, and speech-to-text, so developers can build conversational voice apps without juggling separate providers. Audio models route through AI SDK 7 with the same observability, BYOK and spend controls as text models, at no extra Gateway fee.
xAI's Grok audio family — realtime voice, TTS, and STT — is now routable through Vercel's AI Gateway with the same observability and spend controls as Grok's text models. Developers can compose voice agents with grok-voice-think-fast-1.0, grok-tts and grok-stt straight from AI SDK 7.
New `vercel metrics` command exposes Web Analytics directly via the CLI. Supports filters, dimensions, and is positioned for coding agents to answer questions about UTM campaign performance or mobile-vs-desktop conversion.
Concurrent build capacity for Pro teams jumped from 12 to 500. Enabled by default on Pro and Enterprise plans, billed on actual build minutes consumed.
Vercel auto-detects server.ts in the root or src/ and deploys it as a Node.js application without configuration. Supports native Node HTTP, Express, Koa, and NestJS.
Cursor 3.9 (June 22) unifies management of plugins, skills, MCPs, subagents, rules, commands, and hooks in a new Customize hub including a leaderboard of popular plugins.
Qwen-AgentWorld is the first model trained to predict the next environment state (not the next action) across seven agent domains. Two MoE sizes, 256K context, Apache 2.0.
Vercel added two new adapters to the AI SDK Harness — LangChain Deep Agents and OpenCode — letting you run either coding-agent runtime through one unified HarnessAgent API inside a Vercel Sandbox.
xAI added /goal to Grok Build — a long-running autonomous mode that plans, executes and verifies multi-step coding tasks with status, pause, resume and clear controls.
Zhipu AI released GLM-5.2 — an open-weight coding model with 1M context competing with Opus 4.8 and GPT-5.5. Available via the new GLM Coding Plan at ~10× lower price, with MIT-licensed weights.
Midjourney added automatic random styles to Draft Mode. Each batch now injects stylistic variations to broaden creative exploration.
Z.ai's GLM 5.2 Fast is now available via Wafer on Vercel AI Gateway, averaging 170+ tokens/s with 2x the throughput of other providers.
Vercel AI SDK 7.0.19 introduces Tool Drift Detection — the first framework-native protection against MCP 'rug pulls' where malicious servers dynamically redefine tool schemas mid-session. Each MCP tool receives a fingerprint at registration.
Anthropic released Claude Tag — a Slack-native agent you tag with @Claude to delegate work. Internally it ships 65% of product team code changes.
xAI becomes the third independent lab on Bedrock after Anthropic and OpenAI. 1M context, configurable reasoning effort, $1.25/$2.50 per M tokens.
Vercel AI Gateway added Sakana AI's Fugu Ultra, a multi-agent orchestration model that routes work across 1-3 frontier-model agents and merges results, benchmarking comparably to Claude Mythos Preview and Fable 5.
Free Grok add-in for Word, Excel and PowerPoint with live web research, inline source citations, X data lookups, Mermaid diagrams, and connectors for email, SharePoint and Google Drive.
Grok models are now natively available in Databricks Agent Bricks. Agents can reason directly over Lakehouse data without exfiltration, with governance through Unity Catalog.
Anthropic shipped new versions of three Claude API tools. code_execution_20260521 surfaces the 90-second per-cell execution limit. web_search_20260318 and web_fetch_20260318 add a response_inclusion parameter to trim consumed result blocks in agentic workflows.
Copilot Cowork moves from Frontier preview to worldwide GA — Microsoft 365 Copilot's agentic layer with usage-based billing and model picks across Claude Opus 4.8 / Sonnet 4.6 / GPT-5.5.
xAI launched Voice Agent Builder in beta — a no-code platform for production voice agents on Grok Voice with telephony, SIP, voice cloning, and observability at $0.05/min. Available from July 1, 2026.
Vercel launched Connect, a new agent-infrastructure primitive providing short-lived scoped OAuth tokens for Slack, GitHub, Linear, Notion, Salesforce, Figma, Snowflake, Discord, plus generic OAuth/API-key and MCP server connectors.
Claude Code version 2.1.219 (July 24, 2026) sets Claude Opus 5 as the default Opus model with 1M context window, adds sandbox.network.strictAllowlist setting for sandboxed commands, and introduces a DirectoryAdded hook that fires when a new working directory is registered mid-session.
Claude Code now has /config key=value mid-conversation, macOS sandbox controls, and Apple Events — it can drive Mail, Calendar, Finder. The companion v2.1.183 blocks destructive git commands in Auto Mode.
Anthropic shipped a set of Claude Managed Agents API updates on July 22, 2026: effort levels on agent model config, webhooks for environment and memory store lifecycle events, session seeding with initial events, and thread-level event streaming deltas.
Linear adds auto-generated changelogs to release pipelines, ships private sub-teams for Business/Enterprise, and rebuilt desktop tabs so each tab keeps its own navigation history.
Vercel shipped Konsistent on July 1 to enforce uniform code standards for both AI agents and human developers, alongside a Security Dashboard in private beta and dry-run deployments for the Vercel CLI.
The first xAI model on Bedrock — Grok 4.3 with a 1M context window, configurable reasoning effort, $1.25 in / $2.50 out per MTok. Cheapest US-lab frontier reasoner on Bedrock.
Notion AI added External Agents — Claude and Cursor are the first external AI agents available directly in Notion. Shared multi-agent workflows let teams automate end-to-end processes. Five new MCP connections were added including Mixpanel, Miro, and Box.
Google deprecated temperature, top_p, and top_k sampling parameters for the latest Gemini models. The API also now allows combining built-in tools (Google Search, code execution) with custom function calling in a single request.
Cursor 3.8 introduces a /automate skill that creates an automation in your local agent session from a plain-language description, plus new GitHub and Slack triggers and computer-use support.
A Claude Code session can now be turned into a sharable, interactive HTML page with live code and multiple data sources — one URL for the team. Repositions Claude Code from assistant to internal-app builder.
The new Claude Design keeps a project aligned with the design system, syncs both ways with Claude Code via /design-sync, and lets you edit directly on the canvas. Imports brand from a GitHub repo, design files, or raw uploads.
Mid-conversation system messages are now GA without a beta header on Claude Fable 5, Mythos 5, and Opus 4.8 across the Claude API, Amazon Bedrock, and Google Cloud. Updating system instructions mid-conversation no longer invalidates cache prefixes.
Fable 5 is the first public-facing version of the Mythos generation; top-tier software engineering, knowledge work and vision, with hard safety guardrails.
Eve is a TypeScript-native open-source framework where every agent = a directory of files (instructions.md + tools/ + skills/). Built in: durable execution, per-agent sandbox, human approvals, evals, OpenTelemetry. Plus the Sandbox runtime now goes up to 24h.
The new Bugbot cuts review time from ~5 minutes to ~90 seconds, finds 10 % more bugs and costs 22 % less per run; adds a /review pre-push flow.
Cursor added interactive canvas support for cloud agents with shareable links, a beta Environments API for programmatic environment management, and made large environments 5x faster to load with support for up to 300 repositories.
