# Multi-Agent İlgili Makaleler

HTX Haber Merkezi, kripto endüstrisindeki piyasa trendleri, proje güncellemeleri, teknoloji gelişmeleri ve düzenleyici politikaları kapsayan "Multi-Agent" hakkında en son makaleleri ve derinlemesine analizleri sunmaktadır.

Agents Have Entered the Harness-Driven Era

The article discusses the significance of the leaked Claude Code from Anthropic, highlighting its revelation of advanced Agent engineering practices centered on "Harness" design. Rather than relying solely on model capabilities, modern AI systems now depend on a structured engineering framework—the Harness—to maximize performance. This framework includes six core components: multi-layered System Prompts, Tool Schema, Tool Call Loop (with Plan and Execute modes), Context Manager, Sub-Agent coordination, and Verification Hooks. The Harness enables tighter integration between training and inference, supports long-chain tool execution, and improves reliability through objective verification. It also drives six key training directions: behavior alignment via System Prompt, end-to-end tool-use training, integrated plan-execute training, memory compression, sub-agent orchestration, and multi-objective reinforcement learning. The shift to Harness-driven development reduces the emphasis on pure prompt engineering, favoring instead multidisciplinary talent with skills in AI, backend engineering, and infrastructure. The market is evolving toward more secure, private, and vertically integrated Agent deployments, with "model shell" companies needing either strong infrastructure or deep domain expertise to compete. Claude Code’s leak underscores that future AI advancements will be shaped by engineering architecture as much as by algorithmic innovation.

marsbitDün 10:11

Agents Have Entered the Harness-Driven Era

marsbitDün 10:11

The Use of Humans: Agentic Wallet and the Next Decade of Wallets

The article "The Use of Humans: Agentic Wallet and the Next Decade of Wallets" discusses the evolution of digital wallets in the age of AI agents. It argues that as software users shift from humans to autonomous agents, traditional wallet security models—relying on human confirmation, signatures, and private key management—become inadequate. The core proposition is that Agentic Wallets must serve two masters: humans, who set rules and retain ultimate control, and agents, which require constrained autonomy to execute transactions efficiently. The wallet thus evolves from a simple asset container into a permission and execution system that allows agents to operate within predefined boundaries (e.g., budget limits, approved assets, whitelisted addresses). The article identifies key challenges: current wallets are designed for human interaction, not agentic speed and scale. It outlines four tiers of agent autonomy—from human-controlled to fully autonomous—and emphasizes "bounded autonomy" as the pragmatic near-term solution. A four-layer architecture is proposed: account isolation, permission rules, execution primitives for agents, and governance tools (logging, alerts, veto mechanisms). Critical enabling technologies include standardized Skills (for链上 operations), policy engines, session keys for limited delegation, and audit trails. Current solutions from players like Coinbase, Safe, Privy, and Polygon are noted, but face gaps in portable identity/reputation, unified policy standards, adversarial security (e.g., prompt injection), and cross-chain functionality. The future direction involves a "Wallet Policy Plane" that sits between agent intent and on-chain execution, performing real-time policy checks, risk scoring, and identity verification—akin to Stripe's payment infrastructure. Ultimately, the wallet's role shifts from a front-end gatekeeper to an embedded control layer enabling secure, scalable agentic economies.

marsbit03/22 03:08

The Use of Humans: Agentic Wallet and the Next Decade of Wallets

marsbit03/22 03:08

Post-00s Kid Brother Sets AI Circle Ablaze: 10-Day Project Tops GitHub, Chen Tianqiang Invests 30 Million in a Rush

A Chinese university student, BaiFu, has taken the AI world by storm with two open-source projects—BettaFish and MiroFish—each developed in just ten days using "vibe coding" (AI-assisted rapid development). Both projects reached #1 on GitHub’s trending list. BettaFish, an AI-powered public opinion analysis tool, went viral after a post about it gained traction on social media, attracting widespread attention and investment interest. Chen Tianqiao, founder of Shanda Group, reached out and eventually invested ¥30 million (approx. $4.2 million USD) to support BaiFu’s work. Following this, BaiFu developed MiroFish—a multi-agent AI prediction engine that simulates real-world scenarios to forecast events in finance, policy, and marketing. The project again gained rapid popularity online. This success underscores the rise of the "super individual" era, where a single developer, empowered by AI tools, can quickly build, launch, and scale high-impact projects. Similar trends are emerging globally, with solo founders and small teams creating influential products and companies with unprecedented speed. BaiFu believes that in the AI era, imagination and execution speed are becoming the true drivers of innovation.

marsbit03/09 08:28

Post-00s Kid Brother Sets AI Circle Ablaze: 10-Day Project Tops GitHub, Chen Tianqiang Invests 30 Million in a Rush

marsbit03/09 08:28

活动图片

Indepth Research

Industry News