The article introduces "capability recombination": once a team distills specialized capabilities into permission-shareable agents and tools, individuals can bring their own judgment directly into more kinds of work without mastering every role's details. Using data-service delivery as an example, requesters ask in a group or agent, a dispatch agent assigns the task, specialists and data agents align on definitions, and results return to the original session, cutting manual handoffs. The shift is not that individuals got stronger, but that combining judgment with existing capabilities became cheaper.
Agent
智能体
AI systems that autonomously plan and invoke tools for multi-step tasks, now the dominant source of model API traffic.
Curated Intelligence
LangChain has rebuilt LangSmith around the agent as the core object, launching the LangSmith Agent Platform to unify building, running, improving, and governing agents in one place. The company cites a 2026 step change in agent adoption, with enterprises running coding, customer-facing, and internal productivity agents that need platform-level governance. Developers can start via LangSmith Studio or connect agents they already run.
MotherDuck co-founder Nick Ursa reports that AI has cleared his data engineering backlog for the first time, shifting his job from computing correct numbers to verifying numbers produced by LLMs and agents. The author proposes the emerging role of Data Intelligence Engineer, centered on quality control and decision support rather than ticket queues.
Stripe completed its acquisition of AI model aggregation platform OpenRouter on November 10, with the deal valued at roughly $7.5-8 billion. OpenRouter connects over 400 models from 80+ providers and serves more than 10 million developers, processing about 790 trillion tokens in October 2026, 76% of which came from agent-related API calls. The two will focus on token routing and usage optimization, folding model selection and billing into payment infrastructure.
An AWS blog post explains how agents, skills, and connected data sources fit together in AWS Quick: agents interpret natural-language requests and orchestrate connectors, while skills encode repeatable tasks such as weekly status updates, release notes, and ticket creation into reusable, team-shareable instruction units. It also covers combining the official skill catalog with custom agents, but remains a vendor-bound conceptual and how-to guide without new models or quantitative results.
To address the opacity of AI agent task results, Baoyu proposes requiring agents to self-test and generate screenshots as evidence of completion for human reviewers. The practice borrows from his team's rule that engineers attach feature screenshots to pull requests, and it lets agents catch and fix issues during the screenshot process.
Air Street Capital has released its ninth annual State of AI Report, surveying the past year across research, industry, politics and safety. It concludes that the global frontier has consolidated into a three-way race among Anthropic, OpenAI and Google, while flagging benchmark collapse, agent risks spilling into production systems, and AI turning into a geopolitical lever.
Meta has added Spaces to its Muse agent interface, letting users create separate sidebar conversations around specific projects such as vacation planning, gift tracking or wedding prep, with Muse working toward a goal inside each space. Meta frames these as "custom experiences," but they are essentially topic-scoped breakout conversations, similar in positioning to rival cartoonish agent assistants.
The Verge editor Nilay Patel argues on The Vergecast that AI products do not yet deliver enough value to justify their infrastructure and social costs, that agents remain largely unreal, and that industry spending far exceeds revenue. He contends the AI industry cannot escape its credibility and business problems until products visibly improve users' lives without needing persuasion.
China's National Radio and Television Administration set a target of at least 95% AI penetration across the broadcasting and online audio-visual sector by 2030, alongside more than 150 million new-type terminal applications, as part of its 15th Five-Year Plan. The goal covers AI adoption in content, networks, services, security and regulation, including industry-specific large models and agents, not the share of AI-generated content viewers see.