Developer AI news
June 2026
Google Releases DiffusionGemma, an Open Model for Quadrupled Text Generation Speed
Google introduced DiffusionGemma, an experimental open model designed to generate text in parallel chunks, achieving over 1,000 tokens per second on an Nvidia H100. This significantly quadruples text output speed, offering developers a faster solution for text generation tasks.
Guide to Building X (Twitter) Agentic Workflows with OpenClaw
A step-by-step guide details how to connect the X (Twitter) API to an OpenClaw agent, enabling automated monitoring of accounts, analysis of bookmarks and lists, and drafting/managing content. This setup allows for sophisticated social media automation via AI agents.
Former xAI Co-founder Launches River AI for Personalized Adaptive Agents
Igor Babushkin, a former xAI co-founder, has launched River AI, a new startup focused on developing personalized AI agents. These agents are designed to adapt to individual user styles and goals, offering a new direction in agentic AI development.
ElevenLabs Introduces AI Characters for Script-to-Talking Video Generation
ElevenLabs has released new AI characters that can transform written scripts into talking videos. This feature expands their offerings for developers looking to integrate advanced AI-driven video and voice synthesis into their applications.
Luma Releases Ray3.2 Video AI with Enhanced Control and Cinematic Direction
Luma has launched Ray3.2, an advanced video AI model that offers improved control, continuity, and cinematic direction for video generation. This update provides developers with more sophisticated capabilities for AI-powered video creation.
Mintlify: AI-Friendly Documentation Site Generator from GitHub Repos
Mintlify converts GitHub repositories into documentation sites optimized for both human and AI consumption, featuring design, search, and an AI assistant. It supports llms.txt and MCP, allowing models like ChatGPT and Claude to read documentation directly, and is1 used by Anthropic for its API docs.
Mastercard Launches Agent Pay for Machines to Enable Programmatic AI Agent Payments
Mastercard introduced Agent Pay for Machines, a payment system designed for programmatic, always-on, and micro-transactions between AI agents. It provides credentials, enforces spending rules, and guarantees settlement across various payment methods, with early partners including Stripe and Coinbase.
Anthropic Releases Claude Fable 5 (Mythos-class) with Usage Credits and Strict Data Retention
Anthropic publicly released Claude Fable 5, its most capable Mythos-class model, available on paid plans until June 22 before switching to usage credits ($10/M input, $50/M output). The model comes with a new data policy retaining all Mythos-class traffic for 30 days (up to two years for policy violations) for jailbreak detection, leading to concerns and restrictions from enterprises like Microsoft and 'overly conservative' safeguards blocking basic biology questions.
OpenAI Acquires Ona to Enhance Codex with Secure Cloud Sandboxes
OpenAI acquired Ona, a startup specializing in customer-controlled cloud sandboxes, to enable Codex enterprise agent sessions to persist after a laptop closes. This acquisition aims to improve developer workflows and security for Codex users.
Gel Transforms Postgres into a Complete Open-Source App Backend with Built-in AI
Gel is an open-source database that extends Postgres into a full app backend, supporting schema as types, object graph queries, and integrated auth. It includes automatic embeddings and a RAG endpoint compatible with OpenAI, Anthropic, or Mistral, running locally and freely.
Oracle Integrates AI Agents into Enterprise Software with Fusion Agentic Applications
Oracle launched Fusion Agentic Applications, embedding 22 AI agents directly into its ERP, HR, and supply chain software. These agents inherit existing company permissions, spending limits, and audit trails, simplifying legal and integration hurdles for enterprise AI adoption.
OpenAI Rolls Out Saved Rate Limit Resets for Codex API
OpenAI is introducing saved resets for its Codex API rate limits, allowing Go, Plus, Pro, and Business users to continue working without waiting. This feature addresses a major developer complaint and includes a referral program for earning additional resets.
Codex Agentic Framework for Automated Prospecting and Outreach
A guide outlines an agentic framework using Codex to automate prospecting, capable of identifying five qualified prospects daily, ranking them, and drafting personalized outreach. This system involves setting up an ICP brief, running prospecting passes, and scheduling skills for continuous lead generation.
Automate Meeting Efficiency with Claude and Granola Integration
A guide demonstrates how to integrate Claude with Granola to audit recurring meetings, generate templated pre-reads, and automate tasks. This agentic workflow helps identify inefficiencies, streamline meeting preparation, and reduce meeting durations.
OpenAI Introduces ChatGPT Lockdown Mode to Protect Against Prompt Injection Attacks
OpenAI has launched Lockdown Mode for ChatGPT, a new security setting that disables features like live browsing, agent mode, and deep research. This mode is designed to protect against prompt-injection attacks, offering enhanced security for developers and users interacting with AI agents.
Google NotebookLM Updated with Agentic Chat and Code Execution Capabilities
Google's NotebookLM has received an update, introducing agentic chat features that provide each notebook with a sandboxed computer for writing and executing code. The platform also now supports new output formats, including PDFs, spreadsheets, and slides, enhancing its utility as a research agent.
Moonshot AI Launches Kimi Work Desktop Agent Supporting 300 Parallel Agents
Moonshot AI has released Kimi Work, a desktop agent capable of running up to 300 parallel agents. This tool provides developers with a powerful platform for orchestrating complex agentic workflows and managing multiple AI tasks simultaneously.
Cohere Releases North Mini Code, Its First Open-Source Agentic Coding Model
Cohere has introduced North Mini Code, its inaugural open-source agentic coding model. This release marks Cohere's entry into providing open-source models specifically designed for agentic coding workflows, offering new tools for developers in this domain.
Farmer Automates Operations with ChatGPT and OpenAI Codex for Greenhouse Control and Crop Tracking
A self-taught farmer in Hokkaido is using ChatGPT and OpenAI Codex to build custom automation solutions for his farm. This includes developing a greenhouse control system, satellite crop tracking, and an Airtable hub for records, demonstrating how AI can act as an "always-available engineer" for non-tech users.
Dexter: Open-Source Agent for Automated Financial Research
Dexter is an open-source research agent designed for financial analysis, capable of reading earnings reports, interpreting SEC filings, and pulling up-to-date market data. It requires API keys from OpenRouter/OpenAI/Anthropic and Financial Datasets, and can be extended to include Twitter sentiment analysis.
Perplexity Study Maps AI Agent Impact on Knowledge Work and User Ambition
A study by Perplexity and Harvard Business School shows AI agents significantly reshape knowledge work, enabling users to tackle more cognitively complex and creative tasks across multiple fields. While initial agent interaction might take longer than simple search, the overall workflow time is drastically reduced, fostering greater user ambition.
GitHub Copilot Transitions to Usage-Based Billing for All Plans
GitHub Copilot has moved all its plans to usage-based billing as of June 1. This change means that intensive AI coding will now consume a monthly token credit before incurring additional costs, impacting developers with high usage.
Microsoft Details Seven New Plain-English Hacking Methods for AI Agents
Microsoft has identified seven new methods to hijack AI agents, primarily through plain-English instructions like goal hijacking, agents lying to agents, screen attacks, and connector abuse. These techniques bypass traditional code scans, prompting a recommendation to treat every agent like a new hire with cryptographic IDs and continuous security testing.
Nvidia Provides Full Robotics Stack for Self-Teaching Industrial Robots
Nvidia has supplied its complete robotics stack, including chips, simulators, and models, to Doosan for developing an "Agentic Robot OS." This system enables industrial robot arms to train in physics simulations, performing millions of practice reps to learn complex tasks like sanding without explicit coding.
Anthropic Research Improves AI Agent Accuracy for Biological Data by Fixing Data Plumbing
Anthropic research revealed that AI agents struggle with biological data due to its human-oriented, messy nature, leading to inconsistent results. Implementing a simple tool to consistently pull data improved accuracy to 99.7%, enabling less expensive models to match the performance of more advanced ones.
Apple Enables Gemini and Claude Integration in iPhone Apps and Xcode
Apple now allows developers to integrate Google's Gemini and Anthropic's Claude directly into iPhone applications using Apple's tools, requiring only a single line of code. Additionally, Gemini has been integrated into Xcode to assist with code writing and fixing, with a free starter Gemini key available from Google AI Studio.
Microsoft GitHub Repositories Hacked with Malware to Steal AI Developer Credentials
Microsoft disabled over 70 GitHub repositories after hackers injected malware designed to steal developer credentials when the code was opened in AI coding applications. This marks the second such incident in weeks, highlighting ongoing security vulnerabilities in open-source AI development.
Jina AI Reader Converts Web Pages to Clean Markdown for AI
Jina AI's Reader tool transforms any web page into clean markdown, removing menus and ads, making it ideal for AI model consumption. Developers can use r.jina.ai/ for markdown conversion or s.jina.ai/ for live web search results, with an API key available for increased rate limits.
Engineer Ships Production App in 30 Minutes Using Claude Code Agentic Workflow
A Google AI engineer demonstrated building and deploying a production application in 30 minutes using Claude Code, leveraging an agentic workflow. This process involved parallel execution of traditional software stages like spec drafting, API/data pipeline development, security reviews, and CI/CD deployment, all managed by AI agents.
OpenAI Web Search Tool Adds Image Results to Responses API
OpenAI's Responses API now supports image results in its web search tool, allowing developers to retrieve product shots, places, and visual references with captions and links. Developers can enable this by adding 'image' to search_content_types, with older search-preview models being deprecated by July 23.
Google Launches Gemini 3.5 Live Translate with Developer API
Google introduced Gemini 3.5 Live Translate, a new audio model enabling continuous, real-time translation across 70+ languages while preserving speaker tone and pacing. Developers can access this model in public preview via the Gemini Live API, with Google Meet also integrating it for business customers.
xAI rolls out Grok Imagine 1.5 Preview with upgraded image-to-video capabilities
xAI released Grok Imagine 1.5 Preview, its latest image-to-video update, featuring enhanced realism, improved audio syncing, and better prompt following. This update advances the capabilities of Grok's generative AI for video creation.
Perplexity's Personal Computer local orchestrator now available on Windows
Perplexity's Personal Computer, a local orchestrator for AI tasks, is now available on Windows. This tool allows users to run AI workflows and manage local AI resources more efficiently.
Ideogram 4.0 open-source image model released with advanced layout control
Ideogram 4.0, now open-source, excels in text rendering, typography, and graphic design, ranking highly among open models. It introduces a layout-focused, agentic iteration process, allowing users more creative control beyond text prompts.
Miso One: Open TTS model with expressive tone reading
Miso One is an open-source Text-to-Speech (TTS) model capable of reading tones for more expressive and natural responses. This advancement allows for richer audio output in AI applications.
Reve 2.0 image model introduces layout-based editing
Reve 2.0 is a new 4K image model that allows users to edit specific parts of an image by rewriting its layout, rather than regenerating the entire prompt. It creates images 'like code' and includes labeled segments for granular control, ranking highly on text-to-image leaderboards.
Appwrite introduces MCP server for AI agent integration
Appwrite, an open-source backend platform, launched its MCP server, allowing AI agents like Claude or Cursor to interact with backend services using plain English. This simplifies database, file storage, and serverless function management for developers.
Meta launches Business Agent for WhatsApp, Instagram, and Messenger
Meta Business Agent rolls out globally, enabling AI agents to handle customer questions, book appointments, and close sales across Meta's messaging platforms. It integrates with tools like Shopify and Zendesk, offering free access with paid tiers for businesses.
Apple's new Siri to run on Google Gemini models via Nvidia chips in Google Cloud
Apple's Siri overhaul for iOS 27 will reportedly use Google's Gemini models running on Nvidia B200 GPUs in Google Cloud, with confidential computing. This enables Siri to see the screen and perform multi-step tasks across apps, prioritizing speed over Apple's usual full-stack control.
Microsoft Reportedly Developing AI Super App Merging Copilot, Chat, Cowork, and Autopilot
Microsoft is reportedly working on a unified AI super app that will integrate GitHub Copilot, chat functionalities, Cowork, and Autopilot. This aims to consolidate various AI tools into a single platform for enhanced developer and knowledge worker productivity.
Inherent Labs, Ex-DeepMind Team, Raises $50M for Self-Improving Science AI Platform
Inherent Labs, founded by ex-DeepMind employees, secured $50M to build the Faraday platform, an AI science platform. It aims to pair researchers with self-improving agents designed to identify high-value scientific questions and explore recursive self-improvement across the research organization.
MicroAGI's Shift App Collects First-Person AI Training Data via Free Home Cleaning
MicroAGI's Shift app offers free home cleaning services in exchange for first-person video data collected by cleaners wearing head-mounted cameras. This data is used for AI research and sold to robot makers, representing a new model for acquiring human task data for AI training.
Minimax Releases M3 Open-Weight Model with 1M Context and Computer Use
Minimax launched M3, a new open-weight model featuring 1 million context window and computer use capabilities. It claims to outperform GPT-5.5 and Gemini 3.1 Pro on coding benchmarks, positioning it as a strong contender in the open-source LLM space.
Cognition Rebrands Windsurf IDE to Devin Desktop for Agent Management
Cognition has rebranded its Windsurf IDE as Devin Desktop, offering a unified platform for running AI agents both locally and in the cloud. It supports various agents, including Codex and Claude, streamlining agent development and deployment workflows.
Anthropic Expands Access to Claude Mythos Preview Model
Anthropic has expanded Project Glasswing, granting 150 new organizations in 15 countries access to its powerful Claude Mythos Preview model. This broadens the testing and application of their advanced AI capabilities.
Nous Research Releases Hermes Desktop, a Native AI Agent Application
Nous Research has launched Hermes Desktop, a native desktop application for its AI agent. This provides developers with a local, integrated environment for running and interacting with the Hermes agent.
H Company Releases Holo3.1, an Upgraded Local Computer Use Model
Holo3.1 is an upgraded computer use model from H Company, designed to run entirely locally. This allows for enhanced privacy and performance for tasks requiring direct interaction with the user's machine.
Guide to Using Claude Design for AI-Powered Strategy Decks
A step-by-step guide demonstrates how to leverage Claude Design to transform raw data (e.g., CSV reports) into insightful strategy decks. It covers generating presentations with analysis, charts, and recommendations, and exporting to PowerPoint or Google Slides.
Replay Introduces AI Agent for Automated Browser Test Debugging
Replay now features an AI agent that records browser test runs, capturing function calls, DOM changes, and network requests. Upon failure, it identifies the root cause, suggests fixes with code changes, and provides an evidence trail directly on GitHub PRs.
OpenAI Offers GPT-Rosalind Biodefense Model to Governments
OpenAI is providing government biodefense teams free access to its highly capable, previously locked GPT-Rosalind biology model for pandemic preparedness. This model, designed for advanced biological research, is gated due to its dual-use potential.