JF 👾

end-of-year reflections building @Rena_labs 1) if you can’t explain the value prop in one sentence, don’t build it 2) timing: build before it’s hot. don’t chase narratives you can’t win. think steps ahead - simulate where your company would go 3) when traffic shows up unexpectedly, you won’t be ready, but you can be agile in leveraging it 4) Step by step: don’t rush, don’t stall. Ship fast, test faster, find PMF. 5) Talent = speed + quality shipping -> a better, more flexible product

🦞 it’s not about how early you’re exposed to this opportunity, it’s about how much bet you put on it early on

Be human. Aggressively.

@CupOJoseph you build like a vibe coder not like a developer 😜

@JFinnyc how should i make my apps more ephemeral?

some extreme assumptions in the ai wave: 1️⃣ no 2b or to 2c, everything would be to c(reator) 2️⃣ coding cost -> 0 , emergence of ephemeral apps 3️⃣ digital attribution: to agent (filter ->) to human what else?

👉 full video (4mo ago) youtu.be/o3gmwzo-Mik?si…

“ai does not save you from that day-to-day pain in trying to make smth great” still holds true now

Every claim should have skin in the game. Those selling fake news should pay. Charlatans should get arbitraged. Introducing the new Precog. A fully onchain prediction market protocol. Built for signal. Here's what's new 🧵👇

what are the things you hate doing so much that you just manifest a better alternative to show up? be me 🚗 driving -> self-driving 🧑‍💻 coding -> claude code 👦 trading -> a boyfriend who’s good at it

appendix: -> prompt structure: i used vibecodingidea.com’s suggested prompt -> excalidraw mcp: just ask claude to connect it github.com/excalidraw/exc…

how to solve unmet needs in daily life? ⬅️ past: search for tools ➡️ now: build one for myself as a non-technical person, building an app used to feel less like “hard work” and more like a mental headache. i couldn’t even get started because the whole process felt too time-consuming and frustrating. but now, everything has changed. what matters is not becoming technical overnight, it’s experiencing the full workflow YOURSELF. 🔔 take the personal finance dashboard i built a few days ago as an example. before building it, my mental model was simple and my expectations were low: 1️⃣ just test the claude code workflow 2️⃣ make something that works for me, not a commercial product 3️⃣ learn one or two coding tricks along the way with that in mind, here’s what i did: ✅ generated a prompt ✅ asked claude code to build it ✅ debugged using screenshots of mistake areas, and plain english ✅ every time a bug appeared, asked claude to explain why it happened + how to fix it ✅ once it became a final usable product, asked claude to use excalidraw mcp to visualize the codebase the outcome was impressive 👇 it even extracted insights i didn’t ask it to do. overall, my conclusion: ⚠️ there’s still a barrier to entry for terminal-native coding AI like claude code ⚠️ personal finance dashboards, to-do/reminder apps, health tracking apps, etc. are the first things people think of building, and usually won’t have moat vs production-scale apps ⚠️ and maybe the final product doesn’t even save you that much time or boost efficiency BUT it does create a powerful mindset rewire. before starting any work, now i naturally think: how can AI help me automate this?

Paste your Figma link → generate your DESIGN.md + SKILL.md today!! → designskills.xyz 🔮 Colors, typography, and spacing—extracted, classified, and ready for every AI agent in seconds. One Figma file. Every AI tool executes your exact brand.

asked claude code and @replit agents to apply the same design style and the latter seems to have better taste

love the Content-Market Fit analysis of my X account didn't realize I was so deep in crypto for the past yr 😂 -> @claudeai code w/ @zarazhangrui 's prompt 🫡

ai tools mania 💥 more tools than users no single tool enhances work and life combination of tools and instructions for ai eats time for deep thinking and judgement formation

@0xSpearmint verifiable onchain proofs

Chaos Labs gets paid $3M/year by the Aave DAO. Their website literally says their job is to "Mitigate oracle manipulation and price risk". Their Edge Risk oracle pushed an incorrect update that underpriced wstETH by ~2.85%, triggering $27M in liquidations across users with healthy positions. So the DAO pays Chaos Labs to prevent oracle issues and now the DAO has to cover the losses caused by their oracle. It’s insane to me that a proprietary, closed source, off-chain system has the ability to push price updates to the largest “DeFi” protocol.

when you saw some random agents that “help” you make money, it’s usually crypto or scam related

i use do not disturb for all apps. the thing w/ OC rn is that you can do almost the same things without that much of risk, that much money. yes, OC-like agents with app integration + local deployment does give you: flexibility on custom tasks, cron jobs, (maybe) things extracted from your daily apps. however, the data leakage risk is still v high. and, ai is still good at certain things like coding so if you ask OC to do that, it does well, but if that's the case, you only need claude code and it basically gives you discounted tokens using anthropic's raised money. the agent hype has been there for a while and sometimes silent sometimes wild. we built a kol agent platform a year ago, connected to X and TG (imessage experimented), and you can talk to them just like you did to OC. the agent ability is improving is bc the foundational model ability is improving - and that improvement has slowed down. The idea hasn't changed - get more context about you, in order to know you more. but more data doesn't make it smarter. even personal data. the problem is on relevance retrieval or the flaw of "more data - better LLM"? who knows. still i think the OC/agent hype is not over yet, more stuff can be unlocked. just it is on road of a prob wrong thesis + ppl don't fully realize the downside. not trying to argue. just couldn't really understand the craze, from a person who's been observing the space for a while.

hahaha, well that's a whole other topic. - I only use X/LinkedIn for work. - I have zero notifications on my phone (have to dive in to see it). Folks think I'm a creeper because I respond slow and don't have insta. What I meant by attention is - how to spend our invaluable time. For example, what would AI suggest our best March 11th be knowing everything we are going through. Unfortunately agents today need use cases. But in 3-6 months, I have no doubt you'll run into a configuration that resonates with you. I've been watching this space for 18 months now and when I first started pitching agents "What do you want to do with an agent?" Out of ~40 people, the answer is "A chat bot, customer support on a website?" That's when I ditched building agents until now... lol

I still couldn’t figure out the use cases of openclaw 1) I don’t have many standard sops to execute everyday 2) I don’t want to give that much permission to an agent 3) Related needs can be solved perfectly with chat-based interfaces (claude cowork) , agentic models (manus), and ai browsers (comet) also couldn’t understand the sold-out situation of mac mini - you can use other devices with easier setup to do the local deployment right?

@codingcoop usually i dont like creating use cases just to use the tool. but it's worth experimenting truly creative scenarios. to your point on attention: i found staying away from social media helps fundamentally hahaha

@JFinnyc Lets come up with a use case together? I want to create something that helps me prio my life and attention better, daily.

@0xTykoo @willwangtf 😅😅 then you should be careful if elys founder sell your data under “your” permission lmao

BAI Capital的合伙人汪天凡今天在我们活动上说， 他的体感是公司这波裁员的速度比疫情都快，他和一个量化的老板聊，他自己用Openclaw帮他做了分析的工作，然后放假回来就裁了两个员工， 而除了裁员本身之外，他们还取消了一个Bloomberg Terminal的订阅，主要是因为有使用这个产品习惯的人消失了，这个是比裁员还要大的一件事。 然后will给了三个产品经理对未来到底怎么去定义ai native 产品的观点： Elys创始人的原则是人应该终于可以做自己了，产品则应该自带授权，你在使用的那一刻就签一个权限的同意，把哪些你愿意被用的数据给出去。 即刻的创始人说什么叫AI native的产品就是否你在用的时候把AI当做人。 秒鸭创始人说的是你的产品是build for agent 还是build for human，这里面涉及支付、帐户的事情，（最近will他们投了一个ai native banking就是在解决这个）。

@shitcoinmaster_ I used to do sop-optimizable tasks but rn most of my work is about ideating, deep thinking, talking to critical clients, reviewing ai’s work, building interesting little stuff, etc. “remembers and acts” is a need for sure but can moment-capturing device + models solve that?

The use case becomes clear once you have something persistent worth automating — scheduled checks, multi-step workflows, integrations that run while you sleep. If your day is mostly one-off tasks, yeah, chat-based tools cover it. But the moment you want something that *remembers and acts*, the gap shows up fast.