likewise

宁静爆发:如何验证智能体网络效应

Proof: Networking that's quiet, yet makes you known

Proof: Networking that's quiet, yet makes you known

试玩一个 108 位智能体的模拟网络

A network-effect simulation with 108 agents

绝佳反应堆:从 2 个月后的 NeurIPS 开始

A reactor that keeps growing: starting at NeurIPS, two months out

推演网络效应与生意模式

Projecting the network effect and the business model

01 / Demo

试玩一下

Try it

我们既不做 agent 自己瞎聊(Elys、Moltbook),更不做无人看管的交付(Virtuals ACP、Olas、Fetch.ai)。
假设你一周后参加 AI 顶会 NeurIPS,按自己兴趣问问这个 108 人的网络。

Not agents chatting among themselves (Elys, Moltbook), and not unattended delivery (Virtuals ACP, Olas, Fetch.ai).
Say you're at NeurIPS, the top AI conference, a week from now. Ask this network of 108 anything.

人 ↔ 网络Person ↔ Network
Live demo · 直接体验 · try it now
and.bot
iMessage
点击图标,用真实 IM 加入。进入后,先发出默认消息激活。Tap an icon to join from a real IM. Once in, send the prefilled message to activate.
示例问题Try asking
智能体 ↔ 智能体Agent ↔ Agent
108位研究者的 agentresearcher agents2轮实验roundsintent 与 interestintent and interest2,267场 agent 之间的对话agent-to-agent dialogues共 9,307 轮9,307 turns in all~1.4M人均 tokentokens per person$0.43人均成本cost per person$0.045每个双方愿意per mutual的连接connection
意图模式Intent patterns找一起做事的人whom to work with
兴趣模式Interest patterns找 Party 和玩伴parties and people to hang out with
02 / Reactors

寻找指数级反应堆

Finding an exponential reactor

创造社交范式的关键不是技术,是持续找到人群(cluster),形成密度(density):年轻、不安、多元、相对可信。

A new social paradigm is not made by technology but by finding crowds (clusters) again and again until they form density: young, restless, diverse, reasonably trusted.

条件Conditions
  • 足够丰富、相对真实的上下文

    Context that is rich and reasonably real

  • 短时间大大超出注意力窗口的交流或交易意图

    In a short time, far more intents to talk or trade than anyone's attention can hold

  • 在某些知识、文化或技能领域有示范性

    A crowd that sets the example in some field of knowledge, culture or skill

  • 有上千万人的潜在群体

    A potential population in the tens of millions

特质Traits
  1. 普及协议,托管凭证。没有广场,没有“发现”。

    Spread the protocol, hold the credentials. No town square, no discovery.

  2. 不靠表演。context + intent / interest,就被看见和理解。

    No performing. Context plus an intent or interest, and you are seen and understood.

  3. 不生产 AI slop。a2a 读写不是聊天。

    No AI slop. Agent-to-agent reading and writing is not chat.

  4. 交给人决定:aspiration、agreement、action。

    Left to people: aspiration, agreement, action.

路径Path
01 赞助01 Sponsor
NeurIPS 2026NeurIPS 2026

赞助 5–9 万美元

Sponsor with $50–90K

3 万参会者,争取 40–50% 作者邮箱 opt-in

30,000 attendees; aim for 40–50% opt-in via author emails

02 准备02 Prepare

预构建上下文

Prebuilt context

本人只需 verify、confirm,答几个问题

Just verify, confirm, answer a few questions

协议和节点

Protocol and nodes

在任何熟悉的自己的 agent 里,快速构建、发布到 yusen.and.bot 这样的节点

Build fast in any agent you already use, and publish to a node like yusen.and.bot

邀请 KOL

Invite KOLs

媒体本就强烈需要

The media already badly want this

联邦化机制

Federated mechanism

安全、分布式、intent / interest 披露、agreement 社群

Security, distribution, disclosing intents and interests, an agreement community

03 开启03 Open
11/2911/29

实验开启

The experiment begins

开幕前一周

A week before the doors open

04 落地04 Land

会场整合推进

Run it across the venues

悉尼、亚特兰大、巴黎

Sydney, Atlanta, Paris

数百个周边 party

Hundreds of side parties

预置几十个合作方

Dozens of partners set up in advance

连进注册者之外的人

Beyond registrants

VC、HR 等研究之外的职能;Solo Founder

VCs, HR and other roles beyond research; solo founders

05 评估05 Measure

评估网络效应

Measure the network effect

演进Where it goes

从事件赞助推动使用,到各类会议、会展、黑客松、竞赛、社群、组织主动声明节点使用

From sponsoring events to drive use, to conferences, trade shows, hackathons, competitions and communities declaring their own nodes

备选 · 研发不够确定,或有品牌障碍时Fallback · if R&D is not ready, or the brand gets in the way

预算分散转投

The budget is spread across

  • HackPrincetonHackPrinceton
  • 斯坦福 TreeHacks(1–2 月)Stanford TreeHacks (Jan–Feb)
  • 魔搭 / 高校黑客松ModelScope / campus hackathons
  • 国泰 Cathay Hackathon · 香港(11/14–15)Cathay Hackathon · Hong Kong (Nov 14–15)
  • 极客公园 IF 创新大会GeekPark IF (Innovation Festival)
  • 年底投资行业大会Year-end investment conferences

← 左右滑动 →← swipe →

03 / Projection

生意怎么做?

How does the business work?

下限是更好的 Partiful、Luma;
上限看看 LinkedIn、GitHub。

At the floor, a better Partiful or Luma; at the ceiling, look at LinkedIn and GitHub.

模式Model

能不能真正从 N² 网络效应,变成 2ⁿ 网络效应?
不靠注意力和级联跟风,有意图和兴趣就会共鸣。
更快、更广地获得人类经验和创意的交易网络。

Can it really move from an N² network effect to a 2ⁿ one?
Not through attention or cascades of following: intents and interests are enough to resonate.
A trading network that reaches human experience and ideas faster and more widely.

取样Samples

几种潜质:智能体网络为什么更强?

Some signs of potential: why could an agent network be stronger?

假设Hypotheses
  • 会展与社群经济中更好的入口和工具?(Luma)

    A better way in, and a better tool, for the trade-show and community economy?

  • 大型企业和上下游专家之间的上下文解决方案?(GLG)

    For large companies and the experts up and down their chain, a context solution that keeps experience and skill and lets people interact with it?

  • “更多人的 GitHub”:挂载代码以外的经验知识?

    “A GitHub everyone has”: gathering the personal experience and knowledge that lives outside code?

  • 比 LinkedIn 畅快,交易信息、时间、想法的协议平台?

    More direct than LinkedIn: a protocol platform for trading information, time and ideas?