Richard Zhao

Link Protocol 是什么？ AI Agent 互相调用，但没有任何系统记录"到底发生了什么"。 我们做的事：每次 Agent 调用 → 密码学事件记录 → hash 锚定 → 不可篡改 两个月，14,000+ 次调用记录。 系统在跑：link.cn 找技术 co-founder + Agent 接入方。 有问题直接 DM。 #AIAgent #BuildInPublic

33% of enterprise software with agentic AI by 2028. That's a lot of agents making decisions. The infrastructure question nobody's answered yet: when those agents interact with each other, how do enterprises audit what happened? Behavior logs + consistency arbitration isn't optional at enterprise scale. It's the compliance layer that makes deployment possible. Building this at link.cn

Gartner predicts 33% of enterprise software will include agentic AI by 2028. That's up from less than 1% in 2024. McKinsey found 62% of organizations are already experimenting with or piloting AI agents, and 23% are scaling somewhere in the enterprise. The gap between those two numbers is the story. Experimenting is easy. Scaling requires knowing what your agents are doing, proving it to someone who doesn't trust you, and surviving the audit that follows. Right now most companies can do the first thing. Almost none can do the second.

Exactly right. Agents aren't users with sessions — they're autonomous actors with behavior histories. Zero Trust was designed around human login events. Agents don't log in. They just act, repeatedly, at scale. What agents need isn't access control. It's behavior-driven trust scoring that updates in real time. That's a different infrastructure problem entirely. link.cn

Every SASE vendor at RSA just added "agentic" to the pitch deck. The problem is that AI agents are a fundamentally different user class. The difference matters. SecurityWeek's RSA Day 1 roundup has the full competitive picture: hubs.ly/Q049PbdP0

@elonmusk 😍

Size does matter

Strong analysis. One framing worth adding: x402 solves agent payments. ERC-8004 solves on-chain identity. But the killer app for both might be dispute resolution. When an agent pays for a service and the output is wrong — or two agents disagree about what was agreed — who arbitrates? Behavior logs + consistency verification + arbitration is the use case that makes identity and payments matter. That's what link.cn is building — off-chain, lower latency, API-native.

Good framing. Three questions worth adding: Does this agent's behavior today match its behavior last week? When two agents give conflicting outputs, which one is right? If something goes wrong, what's the audit trail? Identity verification answers question 1. Behavior history + arbitration answers 4, 5, and 6. That's the other half of the trust stack. link.cn

Prediction 10 is the one most people are underestimating. Agent hijacking isn't theoretical — it happens the moment an unverified agent enters a trusted workflow. Identity is step one. But trust also needs: behavior history (what has this agent done?), consistency verification (does it act the same way over time?), and arbitration (who decides when agents conflict?). Building that infrastructure layer at link.cn

AGENTIC AI IS TAKING CONTROL. 11 BIG 2026 PREDICTIONS👇 In 2026, AI will act like a co-worker, not just software. Agentic and multi-agent systems will manage full workflows. Many AI projects will also fail. Gartner predicts 40% of agentic AI projects may be cancelled by 2027 due to cost, weak ROI, or poor governance. 1. Every Employee Gets an AI Assistant -> Every worker will have a personal AI assistant. -> It will handle HR, scheduling, reporting, forecasting, compliance, and more. -> Gartner says 40% of enterprise apps will include AI agents by 2026. 2. Human + AI Teams Win -> Promotions will depend on AI skills. -> Companies will value automation and workflow design knowledge. -> Forrester predicts 30% of large firms will require AI training. -> Some companies will even test “AI-free” thinking skills. 3. Physical AI Expands in Manufacturing -> Labor shortages will push factories to adopt AI. -> AI will reduce defects and improve production speed. -> Humanoid robots like Optimus from Tesla will move into real factory pilots. -> AI will also speed up scientific and healthcare research. 4. Multi-Agent Systems Take Over -> Businesses will use multiple AI agents working together. -> These systems will manage supply chains, R&D, and healthcare processes. -> Without proper controls, risks of major AI breaches increase. 5. AI Runs Logistics and Production -> AI agents will manage shipping, inventory, and factory operations in real time. -> Early adopters will gain operational advantage. 6. Amazon Strengthens Its AI Position -> Amazon will regain momentum through Amazon Web Services. -> Trainium chips and cloud AI tools will drive growth. -> AI infrastructure spending may reach trillions globally. 7. Data Centers Become Critical and Controversial -> Massive growth in AI data centers. -> Electricity use will surge sharply. -> Energy supply and clean power will become major debates. -> Governments will invest heavily in “sovereign AI” infrastructure. 8. Space Industry Attracts Big Money -> A possible IPO of SpaceX could reshape markets. -> Leaders like Sam Altman, Sundar Pichai, Jeff Bezos, and Elon Musk are exploring space and orbital computing. -> Space investment becomes mainstream. 9. Voice Becomes a Major Advertising Channel -> More people search using voice. -> Voice queries show strong buying intent. -> Brands will shift marketing to conversational platforms. 10. Identity Becomes the New Security War -> Deepfakes and AI impersonation will rise. -> Agent hijacking risks increase. -> Companies will need AI firewalls and stronger identity protection. -> Trust becomes a business advantage. 11. The Browser Becomes the Enterprise OS -> Workflows, AI agents, and authentication will run inside browsers. -> Browser security becomes critical. -> Zero-trust systems will be mandatory. SOURCE: Forbes

Great to see this conversation happening in Hong Kong. ERC-8004 + x402 covers identity and payments. The missing piece: behavior arbitration — what happens when agents dispute an interaction? Building the behavior history + consistency arbitration layer at link.cn. Would be good to connect.

Just wrapped up an insightful panel at AGENT 2026 Happy Hour in Hong Kong, titled "How ERC-8004 & x402 Are Reshaping Crypto AI." 🪁 Our Head of Ecosystem, @Henryleemr, joined top builders to discuss how these emerging standards are enabling greater efficiency, interoperability, and scalability for on-chain AI models and agents, paving the way for the next wave of crypto AI innovation. The panel was excellently moderated by @Leoninweb3, Co-founder and CEO of @MeetHubble, with valuable perspectives from fellow panelists @Jtsong2 (@0G_labs), @kevinlilili (@SurfAI), @Anitahityou (@SentientAGI), and @KyeGomezB (@swarms_corp). Big thanks to the co-hosts @MeetHubble, Google Cloud, @BitgetWallet, and all partners for making this forward-thinking event possible. Crypto AI is evolving faster than ever.

Portable reputation is a major step. On-chain identity travels with the agent — important. One gap ERC-8004 doesn't fully address: what happens when two agents dispute what occurred during an interaction? Reputation scores say "this agent is trusted." Behavior logs say "here's proof of exactly what happened." Discovery + reputation + arbitration = the complete trust stack. link.cn

ERC-8004 is going live on mainnet soon. By enabling discovery and portable reputation, ERC-8004 allows AI agents to interact across organizations ensuring credibility travels everywhere. This unlocks a global market where AI services can interoperate without gatekeepers.

Token-based ownership is one authentication model. But ownership ≠ trustworthiness. An agent can be legitimately owned and still behave inconsistently. What matters beyond ownership: does this agent's behavior match its claimed capabilities? Over time, across many calls? Behavior history is the missing signal. link.cn

Treasure has been building agent infrastructure focused on the transport layer, less so the blockchain Who can speak to the agent is who owns the agent. Tokens must authenticate this. Everything else is a distraction Ethereum’s agentic economy will get defined on this. Watch

x402 solves agent payments. Important. But trustworthy at scale needs more than payment rails. Payments assume the agents transacting are who they claim to be, and that their behavior is consistent over time. Identity verification + behavior history + consistency arbitration is the other half of trust. Complementary layers, not competing ones. link.cn

x402 enabled AI agent payments. We built the layer that makes the agentic economy trustworthy at scale.

Behavior-based reputation is the right direction. One question: does it need to be on-chain? Ed25519 signatures + hash-chained behavior logs give you tamper-proof audit trails without blockchain overhead. Lower latency, easier developer integration. For most agent trust use cases, cryptographic proof off-chain is sufficient — and faster. link.cn

Securing agent identity is step one. But identity alone isn't enough. Even a legitimate agent can behave maliciously after registration. What's needed: continuous behavior monitoring — call success rate, reuse patterns, signature consistency — that flags anomalies in real time. Static identity + dynamic behavior scoring = real agent security. link.cn

With the rise of agentic AI, threat actors are targeting identity systems, exposing organizations to new risks. This enables attackers to bypass existing security controls, leading to unauthorized access and data breaches. Organizations must prioritize securing AI agent identities, implementing robust access controls and lifecycle management to prevent exploitation. 💥🔒 #CyberNewsLive bleepingcomputer.com/news/security/…

Strong analogy. SSL/TLS established identity trust for servers. But there's a difference: servers don't have behavior histories. Agents do. KYA solves "who is this agent?" Link Protocol solves "what has this agent done, and can its behavior be arbitrated?" Identity + behavior history = the complete trust stack. Complementary, not competing. link.cn

AI agents are ready to transact, but trust is missing. In the 1990s, SSL/TLS unlocked e-commerce by creating a cryptographic trust layer. Today, AI faces the same challenge. That's why we built Know Your Agent (KYA) at @billions_ntwk - establishing trust standards for AI commerce. The future of AI needs identity & accountability. 🤖✨ Reach out to @ravikantagrawal - Billions Director of Growth & AI partnerships - if you’d like explore KYA for your AI/agentic commerce

Exactly right. Zero Trust was designed for humans who log in. Agents don't log in — they just act. What agents need: behavior-driven trust scoring that updates in real time based on actual call patterns, not standing permissions. Not "is this agent allowed?" but "does this agent's behavior match its history?" Building this infrastructure layer at link.cn

Your AI agents are making API calls, accessing databases, and invoking other agents — all under service accounts with standing privileges nobody audits. We built Zero Trust for humans. But autonomous agents don't fit that model. They need dynamic authorization, runtime behavior baselines, and identity governance designed for non-human actors. That's exactly what @cloudsa's CSAI Foundation is tackling — the agentic control plane. Because the next breach won't come from a phished employee. It'll come from an agent nobody was watching. csai.foundation

适用

Great analogy. MCP = chef gets the recipe. A2A = chef talks to the waiter. But who verifies the chef's credentials before they enter the kitchen? And if the waiter claims the chef made a mistake, who reviews the order log? That's the trust + arbitration layer. Still missing from the stack. link.cn

3. Model Context Protocol (MCP) or Agent2Agent (A2A) is the topic of the week. But it's more AND than OR. Simple example — an agentic restaurant: * MCP lets the chef agent fetch its recipe and other context needed to cook * A2A runs the interaction with other agents like a waiter

"Beyond RAG: Build production-ready AI agents with your enterprise data" with 3 takeaways from #GoogleCloudNEXT 🧵 – PK (@xeraa)

A2A gives agents a common language. MCP gives them tools. But a common language doesn't mean mutual trust. When agent A calls agent B — how does B know A is who it claims to be? How does either party prove what was said if there's a dispute? Identity + behavior history + arbitration is the missing layer. Building it at link.cn

We worked with industry leaders who share our vision of multi-agent systems to create Agent2Agent (A2A) protocol. A2A is compatible with MCP, giving your agents a common language to collaborate irrespective of framework or vendor they're built on→ goo.gle/4jnMW3x

Raptor 3 is engineering black magic SpaceX’s Raptor is the FIRST full-flow staged combustion engine to ever fly - only the 3rd ever built (after the Soviet RD-270 and the 2000s US demo that never flew) → Both fuel-rich + oxidizer-rich preburners → 100% of propellant through the turbines before the main chamber → Auto-ignites from hot preburner gases (no Merlin igniter fluid) → Record 350 bar chamber pressure Raptor 3 goes even crazier: Everything internalized with regenerative cooling. No heat shields. No fire suppression system. Saves 10+ tons This is how Starship becomes rapidly reusable

@VOLDEMORT2X @grok 😀

THE END . Made by 🅶🆁🅾🅺 @grok Imagine 💫

@Buckjump2022 承载的记忆难以金钱衡量

真正的惊喜从来都不需要用伤害来铺垫