OpenAI just updated ChatGPT's deep research mode to pull real-time context directly from internal workspace tools while respecting existing user permissions. The new connectors link with Outlook, Teams, Google Drive, Gmail, and Linear. It is rolling out to Team, Enterprise, and Edu accounts, as well as Plus and Pro users outside the EEA, Switzerland, and the UK.
Anthropic just rolled Claude Code into individual Pro and Max subscriptions, letting developers run the agent directly in terminal and IDEs like VS Code or Cursor without separate API billing. Usage counts toward your shared 5-hour and weekly plan limits, though having an API key set in your environment will still trigger standard token charges.
Epic Games just revealed plans to let creators build voice-interactive AI NPCs inside Fortnite, following the massive success of its AI Darth Vader. Creators will get access to the same safety guardrails Epic used to keep Vader from going rogue, with live demos showing dynamic voice conversations. Why it matters: Bringing unscripted, talking agents into custom worlds could turn static game maps into living playgrounds.
Samsung just teased the next chapter of its Ultra lineup, promising to pack its top-tier specs and foldable-optimized Galaxy AI into a lighter, more compact body. The device promises voice-driven multitasking on the go and adaptive AI workspace features when unfolded. Ultra is finally going folding with smarter onboard models.
Figma just launched the beta of its Dev Mode MCP server, linking design context directly into AI coding agents like Cursor, Windsurf, and Claude Code. Instead of relying on flat screenshots, LLMs can now pull exact components, tokens, and code syntax straight from Figma files. Why it matters: It brings design systems directly into the agentic coding loop.
ManusAI just launched its new video generation feature, turning single text prompts into fully structured, multi-scene video stories in minutes. The tool automatically plans scenes, crafts visuals, and animates the story. It is currently available in early access for Basic, Plus, and Pro subscribers, with wider access coming soon.
AI pioneer Yoshua Bengio just launched LawZero, a non-profit backed by roughly $30M to build guardrails against deceptive autonomous agents. Its upcoming "Scientist AI" acts like an impartial psychologist, outputting probability scores instead of confident answers and blocking actions likely to cause harm. Why it matters: As frontier models gain agency and self-preservation tendencies, we need watchdog systems at least as smart as the models they police.
OpusClip just rolled out OpusSearch, an AI growth agent designed to automatically search, repurpose, and monetize back-catalog videos for creators. Major YouTubers like Mark Rober and Jubilee have already tapped into it over the past three months. Why it matters: it turns forgotten video archives into recurring ad revenue without manual recutting.
ElevenLabs just added Anthropic's Claude Sonnet 4 to its Conversational AI platform to fix the trade-off between speed and deep reasoning in voice bots. The model brings tighter instruction following, long-context memory, and reliable tool calling to real-time speech. Developers can pick it from their dashboard and deploy with a single line of HTML.
Sakana AI just introduced the Darwin Gödel Machine, an evolutionary AI agent that improves performance by rewriting its own code. In tests, it boosted its SWE-bench benchmark score from 20.0% to 50.0% without human intervention. Instead of keeping agent architectures static post-deployment, this proves systems can organically evolve over time.
Perplexity just launched Perplexity Labs for its $20/month Pro subscribers, moving beyond instant search results to tackle heavy-duty research and development tasks. The tool can spend 10 minutes or more scouring the web, running code, and building interactive web apps, spreadsheets, or dashboards from simple prompts. Why it matters: Perplexity is pushing beyond the "better Google" label into full-on autonomous digital work.
ManusAI just launched Manus Slides, an AI tool that creates complete, structured presentations from a single prompt. Users can click to tweak details, then export or share the deck. It is designed to kill the chore of building boardroom or classroom decks from scratch.
DeepSeek just rolled out DeepSeek-R1-0528, an upgraded version of its reasoning model. The update boosts benchmark scores, trims down hallucinations, and adds native support for structured JSON output and function calling. Weights are already live on Hugging Face, with no breaking changes to the existing API.
ElevenLabs just added true text and voice multimodality to its Conversational AI platform. Agents can now process spoken audio and typed text at the same time, letting users speak naturally and switch to text when spelling out tricky info like emails or card numbers. It is a smart fix for pure-voice transcription headaches.
Factory just opened general access to Droids, its autonomous software development agents. The platform connects directly into engineering stacks like GitHub, Slack, Linear, Notion, and Sentry, giving agents full system context to independently build production-ready software across both local and remote workflows.
ElevenLabs just launched Batch Calling for its Conversational AI platform, allowing businesses to blast out hundreds of concurrent outbound calls with voice agents. Teams can upload contact spreadsheets, inject dynamic variables like names or order IDs, and route calls via Twilio or SIP trunking. It basically automates call center outreach at scale.
Opera just introduced Opera Neon, an experimental browser built for the "agentic web." It can either browse alongside you or actively take action and finish web tasks on your behalf. Access is currently invite-only via a waitlist.
The Browser Company CEO Josh Miller just explained why the startup stopped developing new features for Arc to focus on Dia, its new AI browser. Miller admitted Arc was too complex for mainstream users and cited growing security needs around autonomous AI agents. Arc will now only get maintenance patches, with no immediate plans to open-source.
Mistral AI just introduced its Agents API, a dedicated framework for building enterprise-grade AI agents. It comes with persistent stateful memory, multi-agent task handoffs, and built-in connectors for sandboxed code execution, web search, and MCP tools. Why it matters: developers can now build and orchestrate complex agentic systems without having to maintain conversation histories or tool integrations from scratch.
Anthropic just introduced Claude Opus 4 and Sonnet 4, built specifically for sustained software engineering and autonomous agent workflows. Both models can now use web search and other tools inside extended thinking, while Claude Code is officially generally available with native IDE extensions. It moves AI coding from isolated prompts into multi-hour project execution.
