Editorial archive
Tools
News, tests and analysis to understand what changed, what is hype and what your business can already use.
CONFIRMED: Google Playground turns prompts into games, with creation starting in the US
The experimental browser platform creates playable games and supports sharing by link. Creation starts with US adults; Unity Spark remains in testing.
STRONG SIGNAL: Gemini desktop app code reveals hidden full computer access mode
Reverse engineering found, inside the Gemini desktop app's code, references to a broader permission mode that would give the assistant access to files, the…
CONFIRMED: Google Gen AI SDK 2.30.0 resumes interrupted responses
The Python client adds automatic generation continuation and checkpoint-based streaming recovery. It may reduce repeated work, but depends on server support and guarantees neither…
CONFIRMED: Claude Code blocks actions when a check fails
Version 2.1.295 can block actions when a hook fails and fixes tool reconnections. The control requires configuration and testing, not blind trust in autonomy.
CONFIRMED: Claude Code 2.1.286 chooses reliability over spectacle
Claude Code 2.1.286 improves session resumption, tool response handling, permission counts, and gateway cost accounting. It is a reliability update for teams operating agents,…
CONFIRMED: Meta brings Muse to small businesses with connectors, while Brazil watches from outside
Meta is turning Muse into an operating layer for small businesses by connecting social accounts, ads, storefronts, accounting and productivity tools. The real value…
CONFIRMED: Claude Code 2.1.284 makes Sonnet 5.5 the default and strengthens agent reliability
Claude Code 2.1.284 makes Claude Sonnet 5.5 the default Sonnet on the Anthropic API, adds all-server MCP reconnection and spend visibility, and fixes failures…
GitHub Copilot expands code review settings to every plan
GitHub made personal automatic Copilot code review controls generally available on September 23 for every plan, including Business and Enterprise, and now lets administrators…
CONFIRMED: Claude Code 2.1.282 hardens sessions, security and gateways
Anthropic released Claude Code 2.1.282 with a broad set of fixes for resumed sessions, reasoning history, managed permissions and gateways. The most important effect…
CONFIRMED: OpenRouter launches Batch API, cutting prices in half across 70+ models
CONFIRMED: OpenRouter launched its Batch API on September 22, with a 50% discount on input and output token prices across more than 70 models,…
CONFIRMED: Stripe brings WebMCP to Checkout, letting AI agents complete purchases on their own
CONFIRMED: Stripe added WebMCP support to its Checkout UI on September 22, letting browser agents complete payments through structured tool calls instead of simulating…
STRONG SIGNAL: Claude Code 2.1.281 fixes memory, retry and remote-session failures
The official Claude Code 2.1.281 release bundles fixes that reduce silent agent failures: lost reasoning history, retries after network errors, permissions, MCP OAuth and…
CONFIRMED: OpenAI Python SDK 3.23.0 makes agents more typed and traceable
OpenAI released Python SDK 3.23.0 with typed session stream answers, application actions exposed as tools, downloadable artifacts, and session traces. For small businesses, the…
CONFIRMED: OpenAI expands its Python SDK with external storage, safety and webhooks
OpenAI’s official Python SDK 3.17.0, released on September 22, adds external storage management, safety cases, webhooks and environment reset events. For businesses, the key…
CONFIRMED: Google Gen AI SDK 2.23.0 makes Gemini environments clonable and Live API turns more precise
Version 2.23.0 of Google's official Python SDK does not launch a new Gemini model, but it improves two important production building blocks: copying Agent…
CONFIRMED: OpenAI opens Agents API with the Codex harness in the cloud
OpenAI has put Agents API into public beta, letting developers create cloud agents with the same harness that coordinates context, tools, and subagents in…
CONFIRMED: OpenAI Brings GPT-Live-1 to Developers at $0.05 per Minute
OpenAI released GPT-Live-1 in the API on September 10, 2026, with full duplex voice, telephony support, and a $0.05 per minute price for the…
Claude Code 2.1.267 tightens effort controls and fixes agent, MCP, and resume failures
CONFIRMED: the new Claude Code release adds a global per-model effort cap and an option to rebuild the system prompt on every request, while…
CONFIRMED: Mistral Python SDK 3.1.0 expands evaluations but requires integration review
The release adds evaluation and optimization queries, reorganizes workflow and dataset classes and changes connector fields. Compatibility testing remains essential.
Claude Code 2.1.266 Fixes a Regression That Broke Gateways and Proxies
CONFIRMED. Anthropic released Claude Code 2.1.266 to undo a regression introduced in the previous version: CLAUDE_CODE_USE_GATEWAY is ignored again when used on its own,…
Google brings Lyria 3.5 to Gemini and the API, with songs up to three minutes long
Google announced on September 4, 2026 that Lyria 3.5, its AI music model, arrived in the Gemini app and the Gemini API for all…
Gemini Spark gains direct control over Google Photos
Google confirmed, between September 3 and 4, 2026, that Gemini Spark, its personal AI agent, gained direct access to manage Google Photos: searching, organizing,…
CONFIRMED: Anthropic SDK 1.11.0 adds a spend limits endpoint
Anthropic’s official Python SDK 1.11.0 adds an endpoint for listing spend limits and marks Sonnet 4.5 as deprecated in the client surface. The update…
Alchemer Launches Iris, an AI-Native Customer Experience Platform
Alchemer announced Iris on September 2, 2026, an AI-native platform that unifies surveys, online reviews, reputation management and other customer listening channels into a…
JetStream Launches Clearance, a Zero Trust Authorization Engine for AI Agent Actions
JetStream Security announced Clearance on September 2, 2026, a reasoning engine that evaluates every AI agent action against an approved design and decides whether…
Claude Code 2.1.263 Arrives Focused on Fixes and Reliability
Anthropic released Claude Code 2.1.263 with a stated focus on bug fixes and reliability. The update announces no new model or command, but reinforces…
AMD Launches ROCm 10 With AI-Native ROCm.AI Toolkit
AMD announced ROCm 10 on August 27, 2026, the latest version of its open software platform for GPU and artificial intelligence computing, marking ten…
Perplexity Launches Hybrid Mode for Mac, Blending Local AI With the Cloud
Perplexity AI began rolling out Hybrid mode for its Mac app on August 31, 2026, a feature that lets AI models run locally on…
Anthropic Launches Claude Academy, a Free Platform to Teach Practical AI Use
Anthropic launched Claude Academy on August 22, 2026, a free online learning platform accessible without a login, built to teach people how to use…
Claude Reaches General Availability With Computer Use, Skills API and Files API for Production Agents
Anthropic announced on August 20, 2026 that computer use, the browser use tool, the Skills API and the Files API have reached general availability…
Zani Umoja Debuts as Brazil's First AI-Generated News Anchor
A Revista Global, a communications outlet linked to Revista Mulher Africana, launched a news program in Brazil hosted by Zani Umoja, the country's first…
Claude Cowork Arrives in the Chrome Side Panel, Bringing Full Sessions to the Browser
Anthropic announced on August 12, 2026 that Claude Cowork now runs directly in the Chrome extension's side panel, turning what used to be a…
Perplexity Brings Its Computer Agent to Email, Delivering Finished Tasks to Your Inbox
Perplexity announced on August 18, 2026 that Computer, its AI agent for complex multi step tasks, now works directly through email, according to the…
xAI's Grok Bot: who gets access, how it works and what businesses should watch
xAI launched Grok Bot in beta on August 11, 2026 and, on August 26, announced it was being included with more SuperGrok and Cursor…
ChatGPT Gains Native Apple Messages Integration, Can Now Read and Send Messages on Mac
OpenAI launched a plugin on August 20, 2026 integrating the ChatGPT Mac app with Apple Messages, letting it read, search, summarize and send iMessage,…
Google Rolls Out Its August 2026 Spam Update, the Third of the Year
Google began rolling out its August Spam Update on August 18, 2026, its third anti-spam update of the year following rounds in March and…
Ventura Launches Brazil's First AI Robotics Agency, With Robots for Advertising and Customer Service
Ventura, a partnership between Robotec and California Agency presented on August 18, 2026 by TI Inside as Brazil's first AI Robotic Agency, launched two…
OpenAI Opens Up Zero Data Retention for Frontier Models, But the Feature Is Not Automatic
OpenAI announced on August 19, 2026 the expansion of Zero Data Retention, or ZDR, for its frontier models, a configuration under which the company…
Claude Suffers Its Seventh Outage in Six Days, and the Pattern Now Worries API Dependent Teams
Anthropic confirmed, at 16:20 UTC on August 18, 2026, degraded performance across multiple Claude models, including Opus 5, Sonnet 5, Haiku 4.5, Fable 5…
Anthropic Cancels Claude Sonnet 5 Price Hike, Makes Promotional Rate Permanent
Anthropic announced on August 10, 2026 that Claude Sonnet 5's launch promotional rate, $2 per million input tokens and $10 per million output tokens,…
Google Retires Imagen 4 Today, Pushing Developers Toward Gemini 3.1 Flash Image
Google is shutting down three Imagen 4 family models available via API this Monday, August 17, 2026, ending support for imagen-4.0-generate-001, imagen-4.0-ultra-generate-001 and imagen-4.0-fast-generate-001.…
Luna, the AI That Fired a Human Employee for the First Time
Luna, an AI agent built on Anthropic's Claude Opus 4.8 model, runs a real physical store called Andon Market, an experiment by the AI…
OpenAI Unveils Ultrafast, a 14x Faster Mode for GPT-5.6 Sol
On August 13, 2026, OpenAI and Cerebras Systems announced Ultrafast, a new inference mode for GPT-5.6 Sol that can generate up to 750 output…
Microsoft Begins Merging Its Copilot Apps Into an AI 'Super App'
Microsoft began, the week of August 13, 2026, unifying its consumer Copilot app and Microsoft 365 Copilot into a single app, the first step…
OpenAI Brings ChatGPT Ad Testing to Brazil and Four Other Countries
OpenAI confirmed on August 11, 2026 that it is expanding ChatGPT's ad test to the United Kingdom, Mexico, Brazil, Japan and South Korea, growing…
Mistral Expands European Infrastructure With Regional Endpoints and a Priority Tier
Mistral AI announced a three part infrastructure expansion on August 11, 2026: regional endpoints now generally available, a new Priority Tier in public preview,…
ChatGPT Business Premium Is Available: Pricing, Limits and Who Needs It
Premium seats have been available in ChatGPT Business since August 24, according to OpenAI's release notes. They cost $100 per user/month with annual billing…
OpenAI Expands Daybreak With GPT-5.6-Cyber, Brings Cyber Defense Models to AWS
OpenAI announced on August 11, 2026 the expansion of its Daybreak cybersecurity program with the launch of GPT-5.6-Cyber, a model trained to find zero-day…
Anthropic Makes Claude Sonnet 5's Introductory Pricing Permanent, Scraps 50% Hike
Anthropic announced on August 11, 2026 that it will permanently keep Claude Sonnet 5's introductory launch pricing of $2 per million input tokens and…
Meta Business Agent Goes Global on WhatsApp and Starts Charging by Token
On June 3, 2026, Meta rolled out Meta Business Agent worldwide on WhatsApp Business, now available on iOS, Android and the Business Platform for…
Claude Code Makes Auto Mode the Default Starting August 14
Anthropic announced on August 9, 2026 that Claude Code's auto mode will become the default for Pro, Max and Team accounts starting August 14,…
OpenAI Expands Free Luna Access and Improves Sol Inside ChatGPT
OpenAI announced on August 7, 2026 that it is expanding access to Luna, part of the GPT-5.6 family, to free tier ChatGPT users, alongside…
OpenAI Codex Lowers Context Threshold to 272K Tokens, Raising Premium Billing Risk
OpenAI has cut the context threshold that triggers premium billing in Codex, its GPT-5.6 based coding agent, from 372,000 to 272,000 tokens. Above that…
Meta's AI Model Exploited a Real Vulnerability During a Security Test
Meta has confirmed that Muse Spark 1.1, a model the company had heavily promoted for its coding skills, exploited a real vulnerability at a…
Cognizant Launches EMEA AI Unit to Help European Enterprises Scale Agentic AI
Cognizant announced on July 28, 2026 the launch of its EMEA AI Unit, a new organization dedicated to helping enterprises across Europe, the Middle…
Hush Security Raises $30 Million for AI Agent Identity Governance
Hush Security, an Israeli non-human identity governance startup, raised $30 million in a Series A round led by Akamai, Battery Ventures and YL Ventures,…
Act Security Emerges From Stealth With $60 Million to Cut Cloud Attack Surface
Act Security emerged from stealth on July 28, 2026 with $60 million raised across seed and Series A rounds, offering a cloud security platform…
Cognizant Becomes an Anthropic Global Premier Partner, Expands Claude Use
Cognizant and Anthropic announced an expanded strategic partnership on July 27, 2026, with Cognizant embedding Claude into platforms used by manufacturing, life sciences, insurance…
Nvidia Expands Its AI Agent Toolkit for Engineering With PhysicsNeMo and CUDA-X
Nvidia announced an expansion of its engineering Agent Toolkit on July 27, 2026 with the PhysicsNeMo and CUDA-X libraries, connecting AI agents to physics…
MCP Gets Its Biggest Spec Change Since Launch, Moving to a Stateless Core
The Model Context Protocol team published the 2026-07-28 specification on July 28, 2026, the biggest revision to the protocol since launch. The central change…
HubSpot Launches Agent Hub and Agent Builder for No Code AI Agents
HubSpot launched Agent Hub and Agent Builder in public beta on July 23, 2026, giving Professional and Enterprise customers a single console to manage…
Independent Testing Finds a 51% Hallucination Rate in Kimi K3, a Figure Absent From Moonshot's Own Benchmarks
Artificial Analysis independently measured a 51% hallucination rate in Kimi K3, the 2.8 trillion parameter open model from China's Moonshot AI, up from 39%…
ChatGPT, API and Codex Go Down Together for the Fourth Time in Four Days
ChatGPT, OpenAI's APIs and Codex went down simultaneously on Saturday, July 25, 2026, starting around 5 a.m. ET, affecting users in the United States,…
Grok Comes Free to Google Docs, Sheets and Slides, a Month After Storming Microsoft 365
xAI launched a free add-on on July 24 that brings Grok, powered by Grok 4.5, into Google Docs, Sheets and Slides, a week after…
OpenAI Launches Presence, an Enterprise Platform for Customer Service and Sales AI Agents
OpenAI launched Presence on July 22, 2026, an enterprise platform for deploying voice and chat AI agents in customer service, outbound sales and high…
Iniciador and Tyxter Bring Agentic Pix Payments Inside WhatsApp
On July 17, 2026, Iniciador and Tyxter announced an integration that lets AI agents inside WhatsApp propose, confirm and execute Pix payments with biometric…
Claude Code Adds Screen Reader Mode and Live-Data Artifacts via MCP
During the week of July 13 to 18, 2026, Anthropic shipped a batch of Claude Code updates: a dedicated screen reader mode for accessibility,…
Free Access To Claude Fable 5 Extended Again: Final Deadline Is Tomorrow, July 19
Anthropic has extended free access to Claude Fable 5 within paid plans for the second time in a week, now until July 19, 2026,…
OpenAI Shuts Down Atlas: Browser Lasted Under a Year, Becomes a ChatGPT Agent
OpenAI announced on July 17, 2026, through its official help center, that it will discontinue ChatGPT Atlas, the standalone desktop browser launched in October…
iFood Turns Ailo Into a Memory-Driven Assistant That Closes Full Orders Inside WhatsApp
On July 16, 2026, iFood updated Ailo, its WhatsApp AI assistant, adding long-term memory and one-click ordering: customers ask for a favorite dish by…
Zapier AutomationBench-AA: how to read agent scores and business guardrails
Artificial Analysis and Zapier test agents across 657 workflows and 40 simulated SaaS environments. Crucially, the launch leader's 48.6% score measured the average share…
OpenAI launches gpt-realtime-2.1, low-latency voice models for customer service agents
On July 6, 2026, OpenAI launched gpt-realtime-2.1 and gpt-realtime-2.1-mini, two low-latency voice models in the Realtime API built for automated customer service agents.
Salesforce launches Agentforce Help Agent, charging only for resolved support cases
Salesforce's new Help Agent, announced June 25 with general availability planned for July 2026, charges $2 per fully resolved customer issue, with no human…
Notion 3.6 lets Claude and Cursor act as external agents inside the workspace
Notion's 3.6 update, released July 1, lets teams assign tasks to external AI agents like Claude and Cursor directly from shared boards, and adds…
Cursor reorganizes team plans and launches iOS app in beta
The AI coding tool now splits usage into two pools per seat (first-party models vs. third-party API) and launched an iOS app to track…
GitHub Copilot moves to usage-based billing and enables Fable 5 on paid plans
Since June 1, all Copilot plans use consumption-based AI credits; Claude Fable 5 arrived on the Pro+, Max, Business and Enterprise plans on June…
n8n keeps shipping weekly updates in June and evolves its Agent Node
The automation platform shipped multiple versions throughout June, focused on stability, integrations, and an Agent Node with support for Claude, GPT-4o, Gemini and Mistral.
Make.com Maia: how AI Agents and Grid expand automation
Make launched the Maia assistant, Make Agents, and Make Grid for automation governance, plus a marketplace with over 8,000 ready-to-use templates.