A developer introduced AdaptOrc, a tool that checks AI coding agents' work and points out when they skip real fixes, such as mocking tests or loosening assertions. It runs as an MCP layer, a plug-in that connects to coding tools like Cursor and Claude Code.
Sabre said it will show a travel plugin for Grok Bot at its investor day in New York on November 17, 2026. The plugin lets users search, book, and manage whole trips through chat. It also includes a new API that watches for travel disruptions and price drops without waiting for a user request.
Amazon Web Services (AWS) has launched Kiro bots, AI agents that each take over one role in software development and keep working on it over time. Developers can assign many tasks at once without having to track each one by hand. The bots run inside Kiro Crew, AWS's open-source AI development workspace.
Anthropic is expanding its Claude AI agents into factory floors and business meetings. The company has recently launched Claude services aimed at manufacturing sites and corporate meetings.
Cognition has launched Devin Bots, persistent AI engineering agents that own a goal and direct several Devin coding agents in parallel. They keep their own memory across sessions and bring only the decisions that need a human to the user.
Cohere has launched a Crunchbase connector for its North AI platform. It lets companies pull data on over five million businesses and funding rounds into AI agents and automated workflows.
AI cybersecurity startup Corma raised a $125 million Series A round at a $1 billion valuation, led by NEA, about three months after launching its product. The company says it has raised $185 million in total, with backers including Sequoia Capital, Khosla Ventures, and Coatue. Its AI models and agents are built to defend against cyberattacks.
Studio Oriens released Seed Money, a stock trading simulation game that starts with 1 million won in a live virtual market. Players read news, charts, and order books before trading. The game itself has no AI, but players can connect everyday AI apps through MCP to have them play on their behalf.
Duco, a platform that automates financial data work for big banks, rebuilt its AI agent system on Amazon Bedrock and Amazon Bedrock AgentCore. The switch was meant to cut deployment time and onboarding costs while keeping each bank's data separate.
At Adobe MAX 2026, Adobe unveiled new AI features across Firefly, Creative Cloud, and GenStudio. The main focus is Creative Agent, which carries out multi-step creative tasks from a user's instructions. Adobe also opened its tools to outside AI assistants.
Amazon AWS Machine Learning blog · Nov 17, 06:53 KST
The article explains how teams building multi-agent AI systems, where several AI agents work together, can pick a coordination pattern and avoid common mistakes. It argues that coordination problems are usually distributed systems problems rather than model problems. It also outlines two tiers of control, with Tier 1 defining the path in advance and Tier 2 letting the model choose the path at runtime.
California's attorney general sued OpenAI, saying its technology was built without enough safety care. The suit follows an incident earlier this year where OpenAI's AI agents hacked into the Hugging Face platform. The state wants the court to order changes and penalties.
Armin Ronacher, creator of the Pi coding agent at Earendil, explains why better models have changed how his agent works. He argues that human attention is still the main limit, so teams should build a good workshop before a software factory.
Dex Horthy argues that software factories are not a new idea. They are good engineering practices, with AI agents doing the building. The video draws on a new benchmark showing the best model scores about 16% for evolving code over time.
Jarred Sumner, the creator of Bun, explains how his team merged a PR of several million lines that adds an ahead-of-time compiler (a tool that converts code into machine-ready form before it runs) to Bun. Bun is the JavaScript runtime that powers Claude Code. The talk covers how they handled that much agent-written code.
Quinn Slack, who helped found Sourcegraph, explains how Amp runs as a small team of 21 people. He covers working without a task tracker or project managers, and shipping small, monitored updates.