B

Trading

[aix阿乐 @Leepriest1990]

有个数据挺震撼的：一个没技术背景的设计师，零广告、零粉丝，靠给中小企业装AI Agent，6个月做了21个客户。

他的商业模式很聪明：一次安装费+每月Agent管理费+老板陪跑辅导。核心产品不是Agent本身，是教会老板怎么用AI。

想做AI生意的人，别盯着模型了。中小企业老板不关心你用的GPT几，只关心能不能帮他省一个人、多接几单。把AI包装成服务而不是技术，这才是变现的钥匙。

[Fred-New @Fred834567]

其实大环境越差，越要反其道而行之

1、有人在国内闲鱼和拼多多低价收二手iPhone、华为旧机，简单刷机+换壳后，挂到东南亚Shopee和Lazada卖。
单机赚30-80美金，一个月出货900多台，纯利轻松过几万刀。

2、有人刷海外设计师平台（Behance、Dribbble），把欧美最新潮流配色+穿搭手稿下载下来，用AI翻译成中文+排版优化。
然后打包成《2026春夏欧美风尚素材库》，在小红书和视频号卖149元，月5000+份，复购率高达65%。

3、有人在闲鱼卖Codex换肤，零成本月入过万。别人都在烧Token写代码，有人专门在闲鱼卖Codex桌面端皮肤定制——从二次元到财神爷，从王者荣耀到日本女优，统统能改。
售价66到99元不等，销量破千单的商家不在少数。不写一行代码，靠信息差和审美赚AI工具的第一波红利。

4、有人把奶茶DIY做成一门副业。喜茶、霸王茶姬上线手绘杯贴功能后，闲鱼上催生出“代画杯贴”产业链。标价1元打包全套主题杯贴的商家已售超2000单；
还有人只提供“代上传服务”，收费5块，卖了1700多单，光这一项赚了十几万。

5、还有人天天盯Kickstarter和Indiegogo上还没量产的极简厨房小工具，确认爆款后立刻去义乌和1688找代工厂开模，
海外原品售价$199，还没发货。国内首发¥699，月销5000单，单月流水¥350万。货还在集装箱里，钱已经到账了。溢价4倍，差价就是纯利润。

6、有人在小红书卖AI Skill，0成本爆卖40多万。有教育博主做了AI英语外教Skill，定价68.8元，售出2000+份，创收14万+。

这个世界，96%的财富都藏在信息差里。

中国14亿人，9亿人没坐过飞机，信息差大到离谱。

欧美、日本、东南亚每天都在产生海量新趋势、新产品、新内容，而国内大部分人根本看不到，也不会用。

这就是普通人最应该抓住的机会。

AI时代最值钱的能力，不是写代码，也不是做视频。是你先看到，然后告诉别人。

9亿人没坐过飞机，意味着至少有9亿人还活在信息孤岛里。你要做的，就是成为那道桥。

大环境越差，越要反其道而行之——
别人收缩，你就去放大信息差；
别人抱怨，你就去吃信息差的红利。

[Rowan Cheung @rowancheung]

Reminder to go touch grass

A scientist in Japan demonstrated that the air inside a forest strengthens your immune system for a full month

The study is incredibly fascinating:

Qing Li, an immunologist at Nippon Medical School in Tokyo, took 12 men on a 3-day forest trip. These were slow walks with no breathing exercises and no set route.

Then he tested their blood and found a surge in natural killer cells -- the immune cells your body uses to fight cancer and virus-infected cells.

Remarkably, the effect didn't fade when they went home, it was still measurable 30 days later.

A control group that walked the same distance through a city showed no change at all.

Pine, cedar, and cypress trees release compounds called phytoncides to defend themselves from insects and bacteria. You breathe them in, and Li believes they switch your immune cells into high gear.

He estimates one forest walk a month may be enough to keep those cells elevated all year.

The best medicine might not even need a prescription -- just a walk among the trees.

[OōEli.eth @ooeli_eth]

Reddit上一个年轻人发问：

我很焦虑，害怕以后自己过上平庸的人生，该怎么办？

有一条高赞回复是：

“我总觉得自己注定要成就一番大事，所以不敢随便选择一条普通的人生道路。”

这种想法看似雄心勃勃，实际上很容易让人陷入停滞：因为每一次选择，都像是在承认自己只是普通人。

后来才明白，真正优秀的人往往没有突然找到所谓的“天命”。他们只是选择了一件有难度的事，持续投入十年，把它做到极致。

你的独特性来自你为一件事投入的深度，而不在于你最初选择了什么。

一个把普通工作做到极致的人，远比一千个等待命运暗示的梦想家更耀眼。

人生的意义，很多时候不是想出来的，而是在行动中逐渐显现的。

真正需要警惕的，从来不是平庸，而是长期停在原地。

[Alessandro Palombo @thealepalombo]

I'm Italian. My wife is Chinese, born in Beijing.

Many of you have asked how to visit China for the first time. So we sat down and prepared this 101: which cities, in what order, and how to do a first trip properly.

We are both international. She studied in the US and worked in Hong Kong, then Europe. I've lived in more countries than I can count. We know well what most of you will look for in this trip: this is the advice we give our closest friends.

The short version: start with Beijing and Shanghai, leave Shenzhen for a second trip, and don't treat China as a one-week destination. And mind the weather: avoid peak summer and peak winter.

The good news: you are going to luxury you have probably never seen, better than Dubai, at a fraction of the cost.
Thread 🧵

[得否 @wangdefou]

再给大家分享一个信息差：

如果你还没有找到很合适的AI项目，可以去做1对1私教

鸡娃的家长就是你的客户，我这就刚接了一个这个业务

只要你思路打开，服务对象可以是小学生、初中生、高中生、大学生，任何想要1对1线下教学学习AI的人

[Jesse Olson @JesseOlson]

Assuming $BTC has two more bull runs with diminishing returns.

Tenjo projection:
• 2026 bottom $40K-$45K
• 2029 top $215K-$230K
• 2030 bottom $80k=$85K
• 2033 top $315K-$330K
• 2034 bottom $125K-$130K

If the market boss's give us two more bull runs with diminishing returns, then buying this bottom & holding through to 2034 bottom would produce about 25% return per year.

1. Models break and this one will adjust over time.
2. Talking heads have been calling $1M Bitcoin for the past two cycles, they will call it for the next two cycles as well.
3. Something breaks and no more bull runs.
4. HODL'ing continues to get less attractive
5. Still great for swing trades to accumulate more

Do you think we see two more bull runs?

[rari @0xwhrrari]

Google just dropped a free 1-hour course

On building complete agent systems from scratch:

07:02 - Build your first AI agent
13:34 - Give every agent a specialized skill
23:08 - Engineer the right context
32:56 - Connect agents into graphs
54:11 - Run 1,000 agents in autonomous loops

One agent saves you an hour

A thousand agents replace the team you cannot afford to hire

Most people are still learning how to prompt one chatbot

Google is already teaching how to build entire agent systems

Agent → Skills → Context → Graphs → Loops

Single agents are the old workflow

Agent swarms are the new one

This free 1-hour course is worth more than most paid agent engineering programs

Bookmark it before everyone starts building the same thing

Then build your first agent with the guide below

[Lunar @LunarResearcher]

https://t.co/eflBjoVyV2

[六时夜 @Sixtimenight]

这个人从小白学习AI两个月就找到FDE的岗位，看看他是怎么做到的？

以下是他分享的五个习惯：

第一，他每天会把同一个问题分别拿去问 GPT、Claude、Gemini、豆包和 Kimi，再比较不同模型的回答。

第二，他几乎一直在做东西。

个人网站、AI 小助手、HTML PPT、Obsidian 加 Claude、vibe coding 小工具。

AI 公司招人时看重的也是这一点，你真正做过什么是最有说服力的。

第三，他很早就开始研究公司 JD。

先看市场需要什么，再决定自己补什么。JD 里出现 Coze、Dify、Python、剪映、Figma，他就按需求去学。

第四，他定时参加线下交流会。意识到自己闭门造车，开始主动变化，和招聘方的需求也会更接近。

第五，他一直在主动接触外部信息。很多判断，恰恰是在和别人交流、看真实项目的过程中慢慢形成的。

总而言之，这个小白的成长经历非常励志。看下来，他一边做东西，一边观察市场，再不断调整自己的方向。

[Suraj Sharma @suraj_sharma14]

CANCEL your weekend plans.

You NEED to:

• Build a RAG system that cites sources with page numbers
• Implement hybrid search (dense + sparse) for better retrieval
• Add reranking with cross-encoders for top-10 accuracy
• Set up chunking strategies (500 tokens, 50 overlap minimum)
• Build query expansion for better recall
• Add metadata filtering for scoped retrieval
• Implement citation grounding to prevent hallucinations
• Create an eval harness with 50+ golden test cases
• Track retrieval metrics: hit rate, MRR, NDCG
• Add fallback to web search when confidence is low
• Build query rewriting for ambiguous questions
• Implement parent document retrieval for context
• Add embedding caching to reduce latency 80%
• Use colbert or late interaction for better accuracy
• Build a RAG dashboard showing retrieval quality
• Test with adversarial queries that should return nothing
• Document your chunking strategy and why it works
• Benchmark against naive RAG and show improvement

You have way too much to do.

Bookmark & Repost.

[mousepotato @iluciddreaming]

范冰把一本二十多万字的新书免费开源了。

《前置部署工程师》
讲的是 AI 时代怎么把「产品能做的」和「客户需要的」焊在一起——FDE。

全书在线读，一分钱不收。

[知识猫AI实验室 @GeekCatX]

兄弟们，我也来说一个暴论：
闲鱼正在变成普通人的 AI 接单大厅。
2026 年上半年，闲鱼 AI 服务订单 981.6 万单，同比增长 157%，近 500 万人付费购买。
更夸张的是：
AI 编程与建站：增长 1732%
AI 漫剧：增长 1425%
AI PPT 与办公：增长 264%
这组数据说明，用户根本不想学 AI。
他们只想花钱买结果。
所以别再卖什么“1000 个提示词”“AI 入门资料包”了。
直接卖具体交付：
把需求做成一个能访问的网站
把故事做成一条 AI 漫剧样片
把文档做成一份能汇报的 PPT
普通人现在最该练的，不是精通所有 AI 工具，是：
找到需求 → 用 AI 降低成本 → 按时交付 → 收到钱。
24 小时测试方法也很简单：
搜一个细分关键词，拆 20 个同行链接，做 3 个真实案例，上架 10 个不同场景的服务。
先把第一单跑通，再买会员、学工作流、研究商业模式。
注意版权、隐私、虚假承诺和平台交易规则。AI 能把交付成本打下来，但售后这顿毒打，谁也替不了你。

[Chris Pisarski @chrispisarski]

this is how we built all of these sales/GTM workflows in-house with claude code

step 1) connect claude to all of your tools, these are the ones we use internally:

- Hubspot MCP (CRM)
- Slack MCP (team comms + alerts)
- Google Workspace CLI (docs, sheets, etc.)
- Fathom MCP (call recordings)
- Crustdata MCP (people and company data)
- Instantly MCP (outbound)

step 2) ask claude to build the workflows for you

for example..

a) ICP / TAM mapping:

- export all closed-won deals from hubspot: company, deal size, sales cycle length,...

- if you don't have any closed-won yet: use best open opps + competitors' customers (scraped from their case-study pages / G2)

- enrich all via Crustdata MCP: industry, headcount, headcount by department, geo, funding stage, tech stack, open roles,...

- ask claude to run this prompt:

"you have ./winners/ (one json per closed-won customer) and ./crm_export.csv (deal size + cycle length)

1. flag every attribute shared by 70%+ of winners, weighted by deal size

2. drop attributes any random b2b company would also match

3. write ./icp.md: hard filters (industry, headcount, geo, funding) + soft signals with weights (dept ratios, hiring, tech) + anti-icp (attributes of wins that churned or closed slow)

4. spawn a subagent to blind-score every winner against icp.md. 80% of winners must score 70+. loop until they do."

b) mail infra + warmup:

- never send cold from your main domain, buy 2-5 alternate domains to start (yourbrand-hq com, tryyourbrand com), 20-40 at scale

- 2-3 inboxes per domain. person-first names (john@, not sales@), no numbers

- dns on every domain: spf, dkim, dmarc + custom tracking domain (CNAME)

- disable open tracking, pixel hurts deliverability, track replies only

- warmup 2-4 weeks before sending anything. start 5-10/day per inbox and then start ramping up

- warmup never stops: after ramp, ~20 cold + ~30 warmup per inbox daily, 40-50 total

- capacity: 1,000 cold/day = ~50 inboxes across ~20 domains

- before every campaign: verify the list (bounce rate must stay under 2-3%) + run an inbox placement test (primary vs promotions vs spam)

- check all prospects that are behind secure email gateways (proofpoint, mimecast, barracuda). throttle or just delete those and set up a campaign on linkedin instead

c) signal-based lists:

1. reverse-engineer which signals made your prospects reply

- export your outbound history: every prospect ever contacted (instantly export + hubspot): date contacted, replied y/n, meeting y/n, won y/n + fathom as context

- for each account, reconstruct what was true ON the day you contacted them via crustdata: headcount delta the quarter before, days since last funding, open roles matching your buyer titles at that time, new VP+ hire in the prior 90 days, posts made, etc.

- ask claude to run this prompt:

"you have ./outbound_history.csv (every account ever contacted: date, replied, meeting, won)

1. for each account, reconstruct the signal state at contact date via crustdata: headcount delta prior quarter, days since funding, open roles matching [titles], exec hires prior 90 days, posts, all relevant signals

2. calculate lift per signal: reply rate with signal vs baseline reply rate

3. calculate each signal's window: median days between signal and the replies it produced

4. read the reply threads per signal and extract the angle that worked

5. write ./signals.md: only signals with 1.5x+ lift, each with: lift, window, the proven angle. everything else gets deleted, not monitored

6. verify with a subagent: hold out 20% of history, check the ranking predicts reply rate on the holdout. loop until it does"

- no outbound history yet: run the same analysis on closed-won instead (what signal states existed in the 90 days before each winner entered pipeline)

- THEN set up a watcher via crustdata that watches for these signals in real-time, and add that company to your sequence in instantly

d) lookalike lists off closed-won:

- rank customers first: ACV × speed-to-close × expansion, minus churn (using connectors)

- for each: reconstruct the company as it was when it bought, via crustdata (headcount, dept mix, funding stage, what they were hiring at that date)

- pull the fathom transcripts from that deal: why they bought, in their words, and the proxy for that pain (e.g. "drowning in manual prospecting" → SDR headcount growing with no ops hire)

- ask claude to run this prompt on every new closed-won:

"1. check the customer ranks in the top third (ACV × close speed × expansion, minus churn)

2. reconstruct the company at purchase date via crustdata, not its current state

3. read the fathom transcripts from the deal. extract the buying trigger in their words + the observable proxy (what would this pain look like from the outside)

4. crustdata search: companies matching the at-purchase profile AND the pain proxy. exclude current customers, open pipeline, closed-lost < 6 months old

5. score 1-100 on firmographic match × pain-proxy match. keep 70+, cap at 25

6. verify with a subagent: blind-mix the 25 with 25 random companies that pass basic ICP filters. it must identify the real lookalikes 80%+ of the time. if it can't, the criteria are too generic, tighten and loop"

- you can then push the list into a sheet (google workspace CLI), add it to hubspot, or send it to instantly

e) champion tracking:

- pull every contact on closed-won deals from hubspot: tagged champion/decision-maker + get more context from fathom

- resolve their emails to linkedin profiles via crustdata (batch reverse-email lookup)

- one-time backfill first: enrich all of them, diff current employer vs the deal's account

- create a crustdata watcher on each profile, set up a slack channel to get a notification every time something happens: new funding round, new hire made by that company, web traffic, posts where they mention something specific

- you can then set up claude workflows to react to every signal that is interesting for your industry

ask claude:

"pull fired watchers. for each job change:

1. enrich the new company + the new email of the champion using crustdata

2. score the new company against icp.md. below 70 → log it in hubspot, no alert

3. 70+ → create the company + contact in hubspot, tag champion-landed, assign the AE from the original deal

4. slack the AE: who they are, what they bought, the original deal size, what the new company does + a drafted first touch referencing the history using fathom as context"

these are just some of the workflows you can power by connecting your stack to claude

you can build any dream GTM/sales workflow yourself by just asking claude to go through all of these workflows and build something specific for your industry and company, with all the context it has about you

[Chris Pisarski @chrispisarski]

even Anthropic is hiring BDRs to do outbound sales

one of the best-performing cold email copy formats we use to run our own cold outreach is the "creative idea'" email

instead of pitching your own product, you email a prospect 2-3 hyper-specific campaign/growth ideas for their business, then tie it back to how you'd help them pull it off:

"Hey [name], I had some ideas about how we could target [their customer type] together, here's what I'm thinking," then the idea(s), value prop, CTA

it can be very short:

"Hey Olipop,

Quick idea. We could reach out to gyms and boutique fitness studios that don't stock Olipop yet and get it into their fridges as the "healthy soda" for post-workout

Worth testing?"

or longer:

"Hey Olipop,

Had a few ideas for getting in front of new buyers:

1) Email gyms/studios that don't stock it yet for their fridges
2) Target corporate offices for kitchens + vending
3) Reach out to college campus stores for the Gen Z crowd

We'd run the whole outbound for you. Of course we can do other things too, let me know if any of these are interesting!"

And you can automate the entire workflow using Claude (to come up with the ideas), Crustdata MCP (to feed Claude everything about the company + person, their focus, what they sell, contact info) and the Instantly MCP to send out the emails

[Chris Pisarski @chrispisarski]

every sales rep at Crustdata has an unlimited dinner budget for "high-potential" accounts

personal relationships are the single biggest lever in sales

every expansion, closed deal, renewal is easier once a personal connection has been built

it doesn't matter how good your product is , if the person behind the screen can't trust you at all

and personal connections can only be built when you meet people in-person

this is why we encourage our GTM team to get in front of customers/prospects as often as possible

after one dinner, the goal is the second dinner, a cascading waterfall of meet-ups at their favorite spots

people don't say no to this

and it doesn't have to be a dinner, it can be an in-person office visit, a coffee, a courtside seat at a game, a drink after a conference... whatever it takes to meet in person

the focus is "in-person"

realizing that the more upmarket we move, the more important these relationships are

[Chris Pisarski @chrispisarski]

so anthropic open-sourced their claude usage data?

if i had to start a new company today, i'd spend the next 24h going through all of it

find the fastest growing use-cases getting adoption inside companies

then start an AI services agency that deploys those workflows

for example:

> 20% of US claude usage is content creation and copywriting

1) start a claude-first content agency, help companies build an in-house content machine on top of claude

2) reach out to content leads and CMOs at funded b2b companies:

"hey x, 20% of US claude usage now goes into content creation according to anthropic's new usage report from this month...

we help companies build and deploy a claude-powered in-house content machine across the org. we've done this with a few others already. curious if you'd be down to see how that looks?"

you'll get revenue faster than most "startups" out there

there are probably 100s of workflows like this in the usage data

[Chris Pisarski @chrispisarski]

4 months later and the majority of YC founders i'm speaking with are still trying to hire the person that can do all of this with AI:

- map out the entire ICP and TAM
- build signal-based lists
- build lookalike lists off closed-won accounts
- track every champion who changes jobs and route them as a new account
- watch for any signal (tech stack changes, funding, layoffs, headcount swings on named accounts) and react in real time
- set up the mail infra
- run outbound and auto route qualified leads to sales reps
- build an automated multi-touch sequence across email and linkedin
- build the inbound system end to end: creating content, filtered for the ICP, building out a creator network and distributing across linkedin
- score every inbound against past closed-won
- de-anonymize website traffic, push it into pipeline and add them to the outbound sequence
- run focused AEO efforts
- re-engage closed-lost
- watch product usage for expansion signals automatically / add to a sequence
- analyze sales calls and create feedback loops for all reps
- auto-build a pre-call brief for every meeting on the calendar
- generate the one-pager, ROI model and proposal per deal
- build the expansion play per account
- own the CRM architecture and reporting

...and the list goes on

it's really hard to find one person who can own and maintain all of this

for a long time we had two. they built and ran every workflow above and more, and took us from 700k to millions in revenue

the team has since grown to 3

all of them have 3 things in common:

1) they understand the full sales cycle from prospecting to close
2) they're technical, or at least highly technical in how they think
3) they know how to use claude in the most efficient way (spawning /subagents that verify the output and /loop until the outcome is achieved)

and most of these workflows run for a fraction of what the equivalent sales tools cost, as long as your team builds them internally with claude code and a few external APIs

[aasha @aashatwt]

https://t.co/tyNoXlpedk

[Larus Canus @MrLarus]

🤯 Turned famous literary lines into minimalist visual worlds!

Made with GPT-Image2 using monumental objects, tiny figures, subtle embossing, and large areas of negative space.

4 themes:
1. GRAIN OF SAND — William Blake
2. DREAMS — William Shakespeare
3. WITHOUT FEAR — Rabindranath Tagore
4. BEAUTY / TRUTH — John Keats

Bold from a distance, detailed up close.
Classic words, reimagined as contemporary visual art.

Prompt below 👇