Posts

On April 18, 2026, Peter Steinberger walked onto the TED stage in Vancouver and said something that would land differently depending on where you sit in the AI world: “The lobster is loose, and it’s not going back.” For the uninitiated, OpenClaw — the AI agent framework Steinberger built that has accumulated over 247,000 GitHub stars and fundamentally shifted how developers think about personal AI — uses a lobster claw as its logo. The metaphor ran deeper than branding in his talk, titled “How I created OpenClaw, the breakthrough AI agent.” ...

Alibaba just handed the open-source AI community something remarkable: a model that scores 73.4% on SWE-bench Verified — one of the most demanding real-world software engineering benchmarks — while activating only 3 billion parameters per token during inference. Meet Qwen3.6-35B-A3B, released April 17 under the Apache 2.0 license. The Architecture: Sparse MoE Done Right Qwen3.6-35B-A3B is a Mixture of Experts (MoE) model with 35 billion total parameters, but that number is almost misleading for practical purposes. At inference time, the model activates only 3 billion parameters per token — roughly the compute footprint of a much smaller model, with the knowledge capacity of something far larger. ...

AI agents can now charge purchases to your real Mastercard — no new wallet, no new card, no manual approval for each transaction. As of April 16, Lobster.cash (the agentic payments product by Crossmint) and Mastercard officially partnered to bring Agent Pay to the OpenClaw platform, making OpenClaw the first platform in scope for the integration with over one million active agents in deployment. What the Integration Actually Does Here’s the concrete capability: an OpenClaw agent assigned to book travel, purchase API credits, procure SaaS tools, or execute any other commerce task can now use a consumer’s existing Mastercard to complete those transactions — without the user having to review each individual purchase. ...

The White House and Anthropic CEO Dario Amodei sat down on April 17 for what both sides called a “productive and constructive” meeting — the first official dialogue since a significant rift opened between the Pentagon and the AI safety startup earlier this year. The subject at the center of the table: Claude Mythos Preview, Anthropic’s most capable and most restricted frontier model. What Actually Happened Senior officials including White House Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent met with Amodei to discuss how the U.S. government and Anthropic might collaborate on AI safety and cybersecurity without repeating the trust breakdown that characterized the earlier Pentagon dispute. ...

Anthropic just blew open the design process. On April 17, 2026, the company launched Claude Design — a new product under Anthropic Labs that lets anyone describe what they need visually and watch Claude build it. Prototypes. Pitch decks. Marketing assets. Interactive mockups. All through conversation. This is the kind of product that makes designers rethink their workflow and makes everyone else realize they no longer need to apologize for not being a designer. ...

You built the agent. You deployed the agent. Now: who’s watching it? That’s the question Durham-based InsightFinder has been answering — and on April 16, 2026, the company announced a $15M Series B led by Yu Galaxy, bringing total funding to $35M. Revenue is up 3x year-over-year. Customers include UBS, NBCUniversal, and Lenovo. And the round apparently started with an unsolicited seven-figure deal from a Fortune 50 company that needed what InsightFinder was building before InsightFinder had finished building it. ...

Microsoft’s enterprise AI agent platform just got a clearer pricing model — and the answer to “how much does this cost?” is finally something you can budget against. Microsoft Agent 365 goes generally available on May 1, 2026, with a licensing structure designed to solve one of the core frustrations of the preview period: per-agent pricing that scaled exponentially with deployment size. The new model is simpler: one license per user, all agents included. ...

On April 16, 2026, Perplexity shipped something that lands squarely in the middle of the personal agentic AI conversation: Personal Computer. It’s not a chatbot. It’s not a search upgrade. It’s a Mac-based AI agent that runs in the background, handles your tasks autonomously, and can operate your entire digital life — while you’re asleep. Double-Command to activate. Walk away. What Personal Computer Actually Is Personal Computer is a Mac application that turns your machine into an always-on digital proxy. Once running, it can access your Apple Mail, Calendar, iMessage, local files, and the broader internet — using a team of 20+ AI models and sub-agent architectures to accomplish goals you set in natural language. ...

At TrailblazerDX 2026, Salesforce Co-Founder Parker Harris asked a pointed question: “Why should you ever log into Salesforce again?” Not rhetorical. Directional. On April 15, 2026, Salesforce answered it with Headless 360 — a complete re-architecture of how AI agents access the world’s most widely-deployed enterprise CRM. Every Salesforce capability, every Agentforce workflow, every Slack integration: now exposed as an MCP tool, a REST API endpoint, or a CLI command. No browser. No clicking. No human in the loop required. ...

Anthropic’s latest flagship coding model is here — and it’s a meaningful step up. Claude Opus 4.7 is now generally available across all Claude products, the API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. The headline benchmark number — a 3x improvement in production task resolution — deserves precise context. That gain refers specifically to Rakuten’s SWE-Bench, a real-world production coding evaluation, not the general SWE-bench leaderboard. On SWE-bench Pro (a harder variant), Opus 4.7 scores 64.3%, compared to GPT-5.4’s 57.7%–59.1% range on independent leaderboards. It also posts +13% over Opus 4.6 on a 93-task internal coding eval. ...