# Сопутствующие статьи по теме Agent

Новостной центр HTX предлагает последние статьи и углубленный анализ по "Agent", охватывающие рыночные тренды, новости проектов, развитие технологий и политику регулирования в криптоиндустрии.

AI, Why Does It Also Need to Sleep?

Anthropic's accidental leak of Claude Code's source code in 2026 revealed an experimental feature called "autoDream," part of the KAIROS system, which gives AI a sleep-like cycle. Unlike the prevailing AI agent paradigm of continuous, uninterrupted operation, autoDream operates offline when users are inactive. It processes and consolidates daily logs—resolving contradictions, converting vague observations into facts, and discarding redundant information—while avoiding the accumulation of noise in the limited context window, a phenomenon known as "context corruption." This mirrors human brain function: the hippocampus temporarily stores daily experiences, and during rest, the brain prioritizes and transfers important memories to the neocortex through processes like active systems consolidation. Both systems must go offline to perform memory maintenance, as simultaneous processing and consolidation compete for resources. autoDream differs in one key aspect: it labels its outputs as "hints" rather than definitive truths, requiring verification upon use—a cautious approach unlike human memory, which often constructs narratives with high confidence. The emergence of this sleep-like mechanism suggests that, beyond mere biological imitation, intelligent systems may inherently require periodic rest to maintain coherence and performance. It challenges the assumption that more power and continuous operation always lead to greater intelligence, pointing instead to the necessity of rhythmic cycles in advanced cognition.

marsbit04/07 08:20

marsbit04/07 08:20

Xianyu, the OpenClaw More Suited for Chinese Users

The article "Xianyu: The OpenClaw More Suited for Chinese Users" argues that while OpenClaw (a hypothetical AI agent) has generated significant buzz for enabling AI to perform digital tasks, the Chinese second-hand marketplace Xianyu serves as a more practical and "localized" alternative for many. The author contrasts OpenClaw's high computational costs (token consumption) and complex setup with Xianyu's extremely low-cost, human-powered services. On Xianyu, users can find people to perform a vast array of tasks—from booking hotels and buying tickets to AI subscriptions, hiring help, and even matchmaking—often at a fraction of the cost of using an AI. The platform is portrayed as a massive, flexible ecosystem where unconventional supply meets niche demand, creating a parallel economy with over 200 million monthly active users. The piece draws parallels between the two platforms: OpenClaw acts as a leverage tool using AI and tokens, while Xianyu leverages cheap and abundant human labor. Both systems facilitate a form of arbitrage—OpenClaw with digital tokens, and Xianyu through bilateral transactions between buyers and sellers. Ultimately, the article suggests that for China, with its specific socio-economic context and different attitude towards privacy, Xianyu's human-network-based model is a more effective and accessible "agent" platform than an AI-driven solution like OpenClaw.

marsbit04/07 06:12

Anthropic Tests "Lobster" Conway: Supports Independent UI, Webhook Activation, and Custom Extension Standards

Anthropic is actively developing "Conway," a persistent agent solution designed to provide Claude with an always-on, independently running intelligent environment. It features a standalone UI instance, operating as a traditional chat interface and more of a dedicated workspace. Conway can directly control browsers, connect to external services, and integrate Claude Code for deeper code execution and task handling. It supports Webhook-based invocation, enabling external services or events to trigger automated workflows. Additionally, Conway will support for extensions through the upcoming CNW ZIP standard, allowing developers to build custom tools, UI tabs, and context handlers, similar to an app store ecosystem. This enhancement aims to transform Claude from a reactive conversational tool into a proactive, multi-step task assistant, positioning it as a strong competitor in the AI agent space and advancing the trend toward always-on AI capabilities.

marsbit04/02 06:29

Anthropic Tests "Lobster" Conway: Supports Independent UI, Webhook Activation, and Custom Extension Standards

marsbit04/02 06:29

Claude Code Leak: Unveiling the Five-Layer Architecture and Survival Philosophy of a Top AI Agent

A configuration error in the Bun build tool led to the leak of Claude Code's source code, revealing the architecture and internal mechanisms of Anthropic's AI coding agent. The exposed system consists of five core layers: Entrypoints (routing inputs), Runtime (TAOR loop), Engine (dynamic prompt assembly), Tools & Capabilities (40+ tools with strict permissions), and Infrastructure (caching and remote control, including a kill switch). Key innovations include a biologically inspired memory system with three layers (long-term, episodic, and working memory) and an "Auto-Dream" process that consolidates knowledge. Anthropic’s security measures are extensive, featuring an undercover mode for anonymous contributions, anti-distillation techniques to poison API data, and hardware-level authentication. Future development points to "KAIROS mode"—a always-on background agent capable of autonomous action via webhooks and cron jobs. While the leak offers a rare look into a production-scale AI agent, it also highlights Anthropic’s challenge in balancing transparency and security ahead of its planned IPO.

marsbit04/02 01:07

Claude Code Leak: Unveiling the Five-Layer Architecture and Survival Philosophy of a Top AI Agent

marsbit04/02 01:07

Agentic Workflows for Crypto Research

AI coding agents are transforming crypto research by automating data analysis through natural language. This article demonstrates using an agent (like Claude Code) with the Glassnode CLI to test a hypothesis: whether extreme Bitcoin exchange inflows predict 7-day price drawdowns. The agent autonomously discovers metrics, fetches data, performs statistical analysis, and generates insights. Results show a moderate association—spike days saw ~1.9% greater drawdowns—though the sample is small. A follow-up prompt produces visualizations for validation. The workflow allows rapid iteration and is applicable to various research questions using Glassnode data, requiring only an API key and an AI agent.

insights.glassnode04/01 12:41

insights.glassnode04/01 12:41

Bittensor Goes Left, Virtuals Goes Right: Two Flywheel Paradigms of AI Crypto Projects

The article compares two AI crypto project models: Bittensor's subnet ecosystem and Virtuals' agent-based system. Bittensor uses TAO emissions to incentivize subnets, which compete for daily TAO rewards. Subnets require significant capital to launch (e.g., 871 TAO for a subnet slot) and focus on infrastructure like decentralized computing and AI research. Its ecosystem has high entry barriers, complex onboarding, and targets technical users. Virtuals employs a pump.fun-like model where agent tokens gain value through trading volume, enabling rapid capital accumulation during speculative cycles. It has low entry costs for teams, supports quick iteration via a 60-day trial program, and excels at retail distribution due to its Base chain integration and consumer-friendly concepts. Both models share a liquidity flywheel: demand for subnet tokens boosts TAO, while demand for agent tokens boosts VIRTUAL. Bittensor is infrastructure-oriented with high funding potential, while Virtuals is application-focused, leveraging market momentum for consumer AI agents.

marsbit03/26 09:25

Bittensor Goes Left, Virtuals Goes Right: Two Flywheel Paradigms of AI Crypto Projects

marsbit03/26 09:25

Kimi, Zhipu, Douban Gather at an Encryption Hackathon: What Did AI Developers Build on Monad?

Monad's "Rebel in Paradise AI" Hackathon, held on March 21, 2026, brought together leading LLM providers like Kimi, Zhipu AI, and Doubao to explore AI agent development on its high-performance parallel EVM blockchain. The event focused on three key areas: Agent Payments, Smart Markets, and Application Innovation, with over $40,000 in prizes and resources. Winning projects included OpenAlice (Grand Prize), a locally run trading agent with transparent workflows; Orbit AI (NVIDIA Special Award), a decentralized AI cloud using satellite GPU clusters; and Kimi-swarm, an open-source multi-agent IDE. Other notable winners were Libra, a Git-like system for machine-written code, and Anime AI Studio, a one-stop anime short film generation agent. The hackathon highlights Monad's strategic push into AI, leveraging its high throughput (10,000+ TPS), low latency, and low-cost infrastructure to support autonomous agent economies. Monad's existing initiatives, like the AI Blueprint program and x402 payment guides, further position it as a key infrastructure for AI and DeFi integration, enabling agents to execute transactions, settle payments, and operate as independent economic entities on-chain.

marsbit03/26 02:15

Kimi, Zhipu, Douban Gather at an Encryption Hackathon: What Did AI Developers Build on Monad?

marsbit03/26 02:15

OpenAI Exposes 'Polaris' Project, the '2028 Great Unemployment' May Really Be Coming

OpenAI has disclosed its "North Star" project, aiming to build a fully automated multi-agent AI research system by 2028. This goal, announced by Chief Scientist Jakub Pachocki, includes launching an "autonomous AI research intern" capable of handling specific research tasks by September 2024. The project represents a major strategic shift, consolidating resources and products like ChatGPT and Codex into a unified super-app. The initiative reflects a broader industry trend, with competitors like Anthropic also advancing AI agent integration into workflows, such as through Claude Code on platforms like Discord. However, Pachocki acknowledges significant challenges in safety and controllability, noting that full control over LLMs remains unresolved. Economically, the project could be highly lucrative, with projections suggesting AI agents—including a $20,000/month "research agent"—generating $29 billion annually by 2029. The core implication is profound: if AI can autonomously conduct research, the pace of scientific and AI advancement could accelerate beyond human limitations, potentially reshaping labor markets and research paradigms.

marsbit03/24 02:46

OpenAI Exposes 'Polaris' Project, the '2028 Great Unemployment' May Really Be Coming

marsbit03/24 02:46

From OpenClaw to the History of the Web: When AI Gains Sovereignty, What Remains for Humanity?

From Web1 to Web4: A History of Power and Ownership in the Digital Age This article examines the evolution of the web not as a series of technical upgrades, but as a fundamental shift in power—specifically, who owns data, controls wealth, and wields productive force. **Web1 (Read-Only):** Characterized by one-way communication. Platforms like Yahoo owned all content and users were merely passive consumers, or "traffic," with no digital assets. **Web2 (Read-Write):** Users became content creators, but platforms like Facebook and TikTok established a "panoptic dictatorship." They harvested user data to create immense value, but users retained only usage rights, not ownership, of their digital assets and social presence. **Web3 (Read-Write-Own):** A movement to reclaim digital rights through cryptography and decentralization. It enables true digital ownership (e.g., via private keys) and trustless systems (e.g., DAOs, smart contracts). However, it remains a wild frontier with significant legal and security challenges, lacking a capable "workforce" to realize its full potential. **Web4 (Agent Economy):** The convergence of AI Agents and Crypto. AI Agents (autonomous, task-completing AIs) use Crypto as their native currency for machine-to-machine transactions. This shifts power from humans to algorithms, creating independent AI economic actors. This raises critical legal questions, such as liability for AI errors. The future could lead to two extremes: a utopia of liberated human creativity or a dystopia of extreme inequality if AI power is monopolized by a few. **Survival Guide for Web4:** * **Work:** Become a director and risk-manager for AI, not an executor. * **Invest:** Focus on projects with genuine utility, not hype-driven "air tokens." * **Risk Management:** Prioritize robust legal and compliance frameworks for AI operations. The conclusion emphasizes that understanding the transfer of power and assets is key to navigating the future, urging innovation within the boundaries of regulation.

marsbit03/23 13:32

From OpenClaw to the History of the Web: When AI Gains Sovereignty, What Remains for Humanity?

marsbit03/23 13:32

10 Cold Takes on WeChat's Integration of OpenClaw

WeChat's recent integration of OpenClaw as a plugin has sparked discussions, but its immediate impact may be overestimated. The feature allows users to connect their existing OpenClaw agents to WeChat via a simple setup, but it comes with limitations: no group chat support, no streaming output, and no multi-agent connections. The move is seen as a necessary, though delayed, step rather than a groundbreaking change. WeChat’s approach is elegant—acting as a platform rather than building its own agent, thus supporting various OpenClaw-compatible models. However, the integration doesn’t lower the barrier to effectively using AI agents; it only simplifies access. The real challenge remains in configuring and leveraging agents meaningfully. Long-term, this signals WeChat’s shift from connecting people to also connecting humans and AIs, paving the way for future native AI-agent integrations within WeChat’s ecosystem. But for now, it’s a cautious, incremental update—typical of WeChat’s product philosophy.

marsbit03/23 01:10

10 Cold Takes on WeChat's Integration of OpenClaw