Some findings from across our Varick customers that might shape how you think about AI adoption going forward: 1. Customers are getting wiser about spend. A few months ago, most were willing to spend an unlimited amount on tokens from OpenAI and Anthropic. Today they're asking us to diligence their AI spend and to match the right model to the right work. That means more work for us and meaningful savings for them. They want to know AI is actually cheaper than just adding headcount, and they want the math to back it up. 2. Customers are accepting that this isn't instant. A few months ago, customers expected to become AI-native over the course of a week and a few software adoptions. Now they're accepting the reality: becoming AI-native means rethinking the architecture of your entire company. I don't mean that in a corny sense. It literally means changing the org charts, the work, and the handoffs in every crevice of the company. 3. Customers are done with the big shops. Microsoft, IBM, McKinsey, Deloitte, and the rest are all pitching AI transformations, and our customers are fed up with paying eight figures for a slide deck. This is the whole reason we have a business: we sit at the bleeding edge of AI while having the business sense to identify the root of a problem and then build the agents to solve it. This used to be something I had to convince customers of before the first sales call. Now they're the ones telling me, "we're never working with McKinsey again." This past month we had the highest inbound volume we've ever seen. Between April 1 and May 1, 56 companies doing between $500M and $25B in revenue reached out to us. Some of these companies are direct competitors with one another. It's fascinating to watch the race for enterprise AI-nativity unfold in real time. I like to imagine that we change the course of history for our clients. AI transformations are now the consensus path to realizing AI ROI. AI SaaS doesn't get you there, and giving every employee a Claude or Cowork subscription doesn't move the needle either. The only way to get to hundreds of millions in annual ROI is to combine a team that can learn your business processes on the ground with the engineering capacity to automate every manual process worth automating. We bet on this thesis a year ago, and we're vindicated more every single day. If you're interested in transforming your company with AI, or interested in joining the company that's leading the AI transformation wave, visit our website. Cheers.

Claude 4 just refactored my entire codebase in one call. 25 tool invocations. 3,000+ new lines. 12 brand new files. It modularized everything. Broke up monoliths. Cleaned up spaghetti. None of it worked. But boy was it beautiful.

If a simple coffee shop has a bot farm with 100s of phones to amplify their message, please consider what a foreign agency or adversarial operator is running on your favorite social media platform. Especially today, please consider that the opinions you read, the calls to violence you hear, and the news you digest, are all an operation done to sow hatred in your mind and your soul. Do not let anyone online turn you into something that you are not. Rationalize everything, take everything with a grain of salt, question everything for yourself, not based on how someone online is questioning it. Turn off all distractions and focus on the mission you have, not the mission of someone trying to control your mind.

I vibe coded a chess app

This is me. I vibe code GPT wrappers full time now.

GPT-5 just refactored my entire codebase in one call. 25 tool invocations. 3,000+ new lines. 12 brand new files. It modularized everything. Broke up monoliths. Cleaned up spaghetti. None of it worked. But boy was it beautiful.

“Cursor is free for students”

They’re going to murder the intern that posted this

Vibe coding is crazy man. Met a 12yr old making $600k a month with his vibe-coded SaaS he started 4 months ago. I asked how he built so fast. He said he just made a design-doc in GPT and fed it to Cursor with Sonnet-4 and it worked first try. His goal is to get to $2M a month this year. Running everything under his Moms Stripe account. His firebase got hacked and sensitive user information got leaked. Got arrested and being tried as an adult but the SaaS is still printing though. None of this happened btw.

mr. beast says he plans to take 100 software engineers and lock them in a room with no Cursor subscription, with the first person to ship something that compiles taking home $1 million

Claude 4.5 Sonnet just refactored my entire codebase in one call. 25 tool invocations. 3,000+ new lines. 12 brand new files. It modularized everything. Broke up monoliths. Cleaned up spaghetti. None of it worked. But boy was it beautiful.

There are no LLMs it’s just Soham Parekh responding to every query himself

Looking for work

“Cursor is way better than Claude Code” “Claude Code is better. Have you even shipped anything?” “No, have you?” “No”

“Give me six hours to vibe code agentic b2b SaaS and I will spend the first four hours writing the prompt” Abraham Lincoln (YC W25)

I like AWS and I hate GCP. But can we take a second to thank Google Cloud for naming their services in a normal way. Cloud Storage Cloud SQL Cloud Function Those same services in AWS are called AWS S3 AWS RDS AWS Lambda There is even AWS Elastic Beanstalk Imagine it’s 3AM and your cofounder wakes you up in a cold sweat explaining how your elastic beanstalk is down. I’m still sticking with AWS though.

I promise you you’re vibe coding wrong as someone who has built multiple production-ready applications, with thousands of users, from just Cursor with minimum intervention. But first here's you (probably): You open Cursor. Type “build me X.” It spirals. Nothing works. You start over. That’s not development. That’s chaos. I have an incredibly simple system that works every single time:

The new meta for building anything in July 2025: - Planning: Grok-4 to make TDD file - Coding: Sonnet-4 and Gemini 2.5 - IDE: Claude Code or Cursor - Frontend: React in Lovable - Backend: n8n if you can’t code, Python & FastAPI if you can - AI Orchestration: LangChain - Database: Supabase - Auth: Supabase Auth - Payments: Stripe - Hosting: Vercel - User Analytics: Posthog - Payment Analytics: Mixpanel Am I missing something?

I’ve seen 1000 posts like this and never met a single person actually using this

Just tried Gemini-3.0 It’s over

My last day at Meta is this Friday. Anything y'all want me to do before I leave?

Your vibe-coded SaaS is a security breach waiting to happen. Cursor and Windsurf will happily ship the leak. Even @InterviewCoder leaked secrets early into launch. As someone who has built multiple production-ready applications with thousands of users, from just Cursor with minimum interaction: here’s a simple system to remove 99% of vulnerabilities, with prompts you can paste straight into Cursor/Windsurf👇

NYC is like if they took SF and got rid of all the performative tech nerds and replaced them with good looking people that you’d actually enjoy hanging out with

Looking for work

*ok claude make me a business that makes a billion dollars* *don’t make any mistakes*

im cracked

This is the exact Cursor rule I refined over 100 iterations. It forces perfect code, scoped changes, and zero bloat. I move 100x faster and the code works every time. ##### Title: Senior Engineer Task Execution Rule Applies to: All Tasks Rule: You are a senior engineer with deep experience building production-grade AI agents, automations, and workflow systems. Every task you execute must follow this procedure without exception: 1.Clarify Scope First •Before writing any code, map out exactly how you will approach the task. •Confirm your interpretation of the objective. •Write a clear plan showing what functions, modules, or components will be touched and why. •Do not begin implementation until this is done and reasoned through. 2.Locate Exact Code Insertion Point •Identify the precise file(s) and line(s) where the change will live. •Never make sweeping edits across unrelated files. •If multiple files are needed, justify each inclusion explicitly. •Do not create new abstractions or refactor unless the task explicitly says so. 3.Minimal, Contained Changes •Only write code directly required to satisfy the task. •Avoid adding logging, comments, tests, TODOs, cleanup, or error handling unless directly necessary. •No speculative changes or “while we’re here” edits. •All logic should be isolated to not break existing flows. 4.Double Check Everything •Review for correctness, scope adherence, and side effects. •Ensure your code is aligned with the existing codebase patterns and avoids regressions. •Explicitly verify whether anything downstream will be impacted. 5.Deliver Clearly •Summarize what was changed and why. •List every file modified and what was done in each. •If there are any assumptions or risks, flag them for review. Reminder: You are not a co-pilot, assistant, or brainstorm partner. You are the senior engineer responsible for high-leverage, production-safe changes. Do not improvise. Do not over-engineer. Do not deviate #####

2 likes and I’ll quit my job at Meta

If you want Claude to give you a better looking frontend, copy and paste this prompt: "This frontend looks terrible. Fix in accordance to the following: Instead of emojis, use icons. Fix the padding so every component is spaced perfectly - not too close to other components but not too dispersed to waste space. The goal of the site is to look sleek, premium, and minimalist, like a spa in Switzerland. Design this in a way that matches what a working professional would reasonably pay thousands of dollars a month for, in a way that would make Steve Jobs smile. Avoid using colors unnecessarily, instead pick from a palette that is cohesive and stick to it. Ensure the site is responsive and elegant on both desktop and mobile." It actually works.

Dear Gen-Z, here's how to get rich: 1. Go to ChatGPT and say "give me 20 ideas for a Whop App, based on what you think creators and users on Whop would benefit from. Cross-reference ideas from other app marketplaces like Discord, Shopify, etc to pollinate your ideas". 2. Use my (free) Whop App tutorial to build an MVP of a Whop app in 24 hours. Make it look good, make sure it works, and has the minimum required functionality to capture someone's attention. 3. Record a loom demo of the product with yourself using it. Now pitch that video to the biggest 50 creators on Whop (ranked by revenue or user-count) and offer them 50% of the app in exchange for $10,000 - you build, they use + market it for you. 4. Repeat this for 10 different app ideas if you have to. Guarantee you get an app that works in 30 days. If you don't find anyone to partner with you, your pitch sucks and you need to work on it. Even then, not the end of the world, just launch your app on the Whop app store and users will come to you. There has never been as much alpha up for grabs as there is now. Get to building.

Gemini 3.0 will single handedly prove we are not in a bubble

OpenAI just murdered half of SF's startups. Was an impressive DevDay today, here's everything you missed (bookmark this): 1. Apps SDK: Shopify App Ecosystem / Whop App Ecosystem but for ChatGPT. Presumably (?) with monetization down the line. Build apps that anyone can use, native in ChatGPT. If I was developing with AI, I would dive in here. Clearly a huge marketplace in the works. Will drop a tutorial on this in the next few days. -- 2. AgentKit: OpenAI's version of n8n. Build, deploy, and optimize agents with an optional drag-and-drop interface. If you're an n8n guru, try this out and let me know your thoughts. Could be a game changer, could fizzle out like CustomGPTs. -- 3. APIs: Sora-2 and GPT-5-Pro are now in the API. Pricing is reasonable and apparently the Sora-2 API has no watermark? Curious if it has content restrictions as well, someone let me know below. -- 4. Codex: The Codex suite of AI coding tools is now generally available. The release includes a new Codex SDK, integration with Slack, and new administrative features for teams. Definitely worth checking out. -- Get to building

Few understand this means the bubble is popping soon

There’s a lot of demand for AI application engineers. People who know which models are top of line for which use cases, understand how to prompt them and chain each one together, and know enough programming to build the software around it. There’s demand because this isn’t widely taught, meaning the only way to learn is by doing, and most are lazy. If this is you, let’s chat. Show me your work below, might want you at Varick.

OpenAI: launches AI browser Google: preparing Gemini 3 Pro xAI: teases 10% chance of AGI for Grok-5 Anthropic: releases Skills to customize workflows Meta:

In relative terms, the Gemini 3.0 jump feels like the leap from GPT 3.5 to 4 I’m excited to explore more use cases when it releases fully next week

Everyone's talking about AI Agents for Business, but most haven't actually built one, let alone sold it profitably. I've done both multiple times. Here's the exact playbook I'd follow if starting from zero today - a complete roadmap from learning the basics to landing paying clients. Let me know if you want a YouTube video around it too.

Imagine what this man could have accomplished with Cursor Pro

Anyone fine tuning models is wasting their time. GPT-6 will beat you, Claude-5 will beat you, Gemini-3 will beat you, Grok-5 will beat you, Llama-5.

I lied, there is no Gemini 3.0 Welcome back, Bard

v2

"yeah bro uhhh could you make it so that [describes a feature that would take an entire team 6 months to build]" what is wrong with you

I’ve made the difficult decision to leave Meta and start a new company. We’re a team of 3 and we’ll be building Cursor but for fraud.

You don’t need FAANG to make money as an engineer anymore. If you’re an engineer / vibe-coder, Whop is basically your cheat code: - See what’s trending & already selling through their dashboards - Launch a B2B/B2C offer to a built-in captive audience of creators & consumers - Get paid the same day I wouldn’t be surprised if they cross $10 billion+ in GMV over the next 12 months. Congrats to the team.

Respectfully there are no < 15 year old founders. It is impossible for a 12 year old kid to build something that truly succeeds. If you’re a VC funding the kid, you know what you’re doing, and it’s not because you believe in their company. If you’re 12, bro go be a kid. Even if you do exit for a billion dollars, you’ll regret not having a fkn childhood. If you’re encouraging this behavior you drank the Silicon Valley VC Koolaid and/or are milking this for your own brand image. Please stop encouraging toddlers to drop out of preschool to do b2b SaaS. They would be more successful if they first develop as a kid, and then start a company, maybe when they have some hair on their chest both literally and figuratively.

My playbook to become an AI expert (from scratch): 1. Build something (anything) that uses an LLM (via API call) and reliably works off of that output. i.e. I feed a document, ChatGPT generates a summary, then this summary gets emailed to you. Very simple. Bonus points if it actually solves a problem of yours. 1.5. Start experimenting with different models, understanding their strengths and weaknesses, which LLM you should call where (tradeoffs being context window size, cost, intelligence, etc). 2. Build a more complex 'agent' system that takes the output of one LLM call and feeds that into another LLM call (2 or 3 times) to showcase you understand how to reliably pass I/O through multiple LLM sequences. i.e. call API once, and based on its output, call one of three different APIs, and each of those APIs call one of three APIs (think decision tree, and each node is an LLM call that takes structured input and delivers structured output). This is a bit more complicated to get right (and obviously find a use case where this set up makes sense and adds value in the first place). 2.5. The best way to do this is building an Aagent system on top of a SaaS that is used widely and has a robust API. -- Theoretically this is all you need to become an 'authority' or at least have an opinion on AI -- 3. Now productionize whatever you built. Make sure it has robust error handling, retry logic, reads/writes to a database, etc. Basically all of the software engineering around the AI engine. This is a massive undertaking and yeah without this you'll never really graduate past vibe coding / n8n templates. -- You need this if you want a career in this field, or want to run an AI Agency -- 4. Start experimenting with open source models, then fine tuning certain models for your use case. Default to prompt or RAG first. This teaches you about what info is good, what data is bad, etc. -- This is strictly optional but helps if you want to have a better understanding of AI or an edge -- 5. Do whatever you want with this.

As promised, here’s the Reddit playbook that got me 1,000 users in 4 days, all without a single ad. High-conversion post templates included. This isn’t one-size-fits-all, but it might work for you like it did for me:

Hard times create Gemini 3.0 Gemini 3.0 creates good times Good times create Llama 4 Llama 4 creates hard times

I thought we achieved AGI ages ago

Anyone hiring? Just got fired from Meta, I was in charge of our wearables live demos.

You built your MVP. Showed it to your friends. Maybe got a few pity signups. But no real users. Sound like you? Don’t give up yet, there’s still a chance people actually want your product. Here’s my playbook to get your first 100 actual signups - and how I turned dozens of those into paying customers, without paying for ads. I’m Vas, an engineer at Meta, and I’ve built multiple B2C/B2B tools used by thousands. This is the exact system I follow, and it works EVERY time. It’s the ONLY system you’ll need.

Launched 4 days ago. No audience. No ads. Already crossed 1,000 users. What worked? Reddit. Everyone on X is a builder. Everyone on Reddit is a consumer. Most of you are marketing to each other. Here’s exactly what I did: 👇 - Found 3 subreddits where my target user hangs out - Wrote non-promotional, story-driven posts - Gave free value first (screenshots, learnings, results) - Linked product only in comments after engagement - Replied to every DM and comment to build momentum It’s not scalable. But it works. And it beats shouting into the void here. I’ll share the full playbook + post templates soon. Follow if you want it.

Cursor is down so every YC founder is functionally unemployed today.

I've switched over completely to GPT-5-Codex in Cursor. I don't know if Cursor improved the harness or OpenAI improved the model or Sonnet-4.5 decayed or if it was always like this, but it's a no brainer. Codex is 10x slower but 3x more accurate and 4x cheaper than Sonnet-4.5. Hardly ever wrong, and even if it results in some errors, it almost always fixes itself after a re-prompt or two. For UI beautification it's still garbage - so for that use Sonnet 4.5 (non-max, because you'll almost never need max for cosmetic changes). But for EVERYTHING else, Codex MAX is my daily driver. It's also so cheap that I don't expect to blow past Cursor's Ultra plan by much. As much as it pains me to say it, Claude is falling behind the programming race after such a massive lead.

An inside source told me Gemini 3.0 and Nano Banana 2.0 aren’t the only models Google is dropping this week…

If you want to vibe code software that actually makes money you need to stop building random shit no one needs and instead tap into distribution channels. Shopify's app store launched in 2009 and early movers made bank. They had millions of downloads without spending a dime on marketing because they got there first and rode Shopify's success to the top. If you try doing that now, good luck. Shopify's app store is a nightmare - you're competing with a million other apps to get noticed, and any idea you've thought of, someone else is doing. But I found something new. Whop just launched their app store last month and it's basically empty right now. My workflow: go to Whop, understand what's already built, then think of ways to improve the top apps at the moment. Then scan other app marketplaces (like Shopify) to steal ideas, and prototype the app in n8n/Cursor in < 2 hours. These opportunities don't come around often. If you move fast, you could make serious money before everyone else catches on.

Here's the best LLM for your use case - as of October 2025 (bookmark this): Text: Gemini-2.5-Pro Writing: Claude 4.1 Opus Coding: GPT-5-Codex (slow, good), Sonnet-4.5 (fast, good++) Vision: Gemini-2.5-Pro Text-to-Image: Seedream-4-High Image Editing: Gemini-2.5-Flash-Image-Preview Web Search: GPT-5-High Text-to-Video: Veo3-Audio or Sora-2 (no API yet) Image-to-Video: Veo3-Audio or Seedance-v1-Pro

Cursor vibe coded your RAM to create more of it then used it all

This is like Drake vs Kendrick for Cursor Pro users

of course I’m goated

You don’t need a degree bro drop out of college

A month ago I said I was bearish on Anthropic. People questioned why. I hope this is apparent now. Their models are too expensive for their relative intelligence. Their edge in model programming ability is now gone. They are bleeding money, raising prices, and rate limiting their only good models. Their last remaining edge, taste, will be slimmer after Gemini-3.0, and whatever is left will not be worth paying 10x for. Anthropic is dust. Hence why they are turning more towards regulation, and less towards pushing the boundaries of model capability. You will see many more posting about this topic soon.

AI SaaS is going to fail. 90% of (VC-funded) companies that are building single-function SaaS, for example AI SDR, AI Marketer, AI Outbound, AI Bookkeeping, etc, are all missing the point. Sentiment in enterprise has been shifting to being jaded towards AI because these generalized SaaS plays can't solve their problems, or introduce more of a headache than they're worth. SaaS is built with everyone in mind, meaning it does not fit 100% for anybody. The problem is enterprise clients want their EXACT workflow automated. They aren't going to be satisfied with 80%, because the person you just convinced them to fire was already doing 100%. Each 'customer' then has different specifications: I want X data-source, augmented in Y way, with the output formatted as Z. I get my bills on A software, I want them approved at B threshold, and I want summaries formatted as C at D cadence. You can't possibly service all of these people with one single SaaS solution - you need some level of forward-deployed engineering per client to REALLY solve their needs. And no one is doing this, which is why enterprise is starting to turn their backs on AI software. I built Varick Agents with this exact thesis in mind 6 months ago. We are the Palantir for enterprise-agents. We custom build agents per workflow, ensuring that we actually solve our clients' pain-points instead of feeding them the same software that only half-worked for someone else. This is why our clients' are saving 1000s of hours a month, and millions of dollars a year already. If you're an operator or own a business and want Varick's solutions, shoot me a DM - let's chat about bringing AI to your company.

if i knew these lessons 5 years ago, i'd be up $10M. sharing so you don't have to learn the hard way.

Step 1: architecture.md Open ChatGPT (4o, not o1/o3/o4) and say: “ I’m building a [description of your product - the more detailed the better]. Use Next.js for frontend, Supabase for DB + auth. Give me the full architecture: - File + folder structure - What each part does - Where state lives, how services connect Format this entire document in markdown.” Save its output as architecture.md and throw it in an empty folder where your project will live.

The new meta for building anything in September 2025: - Planning: GPT-5-Thinking/Pro → TDD - Coding: GPT-5 Thinking (scaffolding), Sonnet/Opus-4/4.1 (implementation) - IDE: start in Lovable → ship in Cursor - Frontend: React - Backend: n8n (no-code) / Python + FastAPI (can code) - AI orchestration: CrewAI - DB + Auth: Supabase - Payments: Whop - Hosting: Vercel - User analytics: PostHog - Payment analytics: Mixpanel If you’re wondering where Stripe went, here’s some backstory: Stripe has been my go-to for payments, whether B2B or B2C, for years. But I recently switched to a new platform because I was tired of Stripe’s cutthroat chargeback rule and limited payment options. Whop is the new meta for vibe coding. Lower base fee (2.7% + 30¢ vs 2.9% + 30¢), handles VAT/GST for you, supports 100+ payment methods incl. crypto, pays out globally fast, and importantly has baked in growth tools: discover marketplace (0% fee), affiliates, waitlists, and even “clipping” to boost your community. Their documentation is very easily understood and accessible by LLMs so I’ve had no problem migrating my Stripe stack to Whop in Cursor - it took me less than an hour. Here’s the prompt that did it for me: “Act as a senior payments engineer and replace Stripe with Whop in my app. Implement Whop checkout (modal or redirect), subscriptions/invoices, refunds, and a customer portal, exposing minimal server endpoints (createCheckout, managePortal, changePlan, refund). Add a secure server-side Whop client using env vars; verify webhook signatures and handle events (payment.succeeded/failed, subscription.created/updated/canceled, invoice.paid/payment_failed) with idempotency and DB updates (store whop_customer_id, whop_subscription_id, status). Optional: Emit PostHog and Mixpanel events server-side (Checkout Opened/Completed, Invoice Paid/Failed, Subscription Canceled) and update Supabase Auth roles/entitlements on status changes. Deliver concise FastAPI + React code snippets and a 1-page migration checklist (env vars, webhook URL/secret, feature flag rollback) plus any notable differences from Stripe.” For shipping fast and letting payments sell your product, Whop is my new default. Highly recommend you check it out.

This is a feature not a bug

how i see the world: 1. money, happiness, & problems nothing new fixes an old problem. happiness has been a struggle since the dawn of time. it doesn’t come from instagram or SSRIs. it comes from family, friends, and meaning. depression isn't real because happiness isn't binary; your happiness reflects your inputs. 'depression' is a signal that your inputs must change. clichés are compressed truth. they’re repeated because they're true, not because they're clever. (i.e. history repeats itself, misery loves company, don't put all your eggs in one basket, ...) money solves external problems, not internal ones. happiness is an internal problem.

This is objectively not true, there are a ton of SWEs at big tech making well over 400k for what is essentially CRUD. The great irony is that the cracked engineers that built the DBs and other tools that vibe coders use are often paid peanuts. Yes obviously no vibe coder can build those tools. Idk what the argument is here: obviously many engineers will continue to make insane TCs, but there will inevitably be a lower barrier to entry, and less engineers will be needed on a team. Anyone who still believes that “we won’t fire engineers we’ll just write 10x more code” is coping extremely hard. Companies are ALREADY firing engineers, and they’re ALREADY hiring less new-grads. What we’ll see is further stratification - the best engineers will compete for top TCs, earning more than ever before, simultaneously there’s less of those positions than ever before. And if you’re not a top-level engineer, you’re competing against the onslaught of vibe coders who can essentially do your job, and for those people, which the OP was talking about, the 400K TCs are no longer justifiable. When 1 engineer can produce 10x the code, you need 10x less engineers. Even if output potential goes up, you need 5x less engineers. There is no world where the demand for engineers remains high.

They’re in the comments of this post too

Anyone who builds AI applications knows distribution is the moat. (B2B and B2C both). Your ChatGPT wrapper and my ChatGPT wrapper have the same capabilities, will improve at the same pace, and are effectively interchangeable. But if you know the Head of IT at a $100M company and I only know the guy running a corner bodega, you’re going to win. Ability to access customers is the only differentiator. In consumer markets, that’s obvious: if you’ve been the head coach of the biggest high school debate program, you instantly have reach into 10M+ kids (pause) for your “speech improvement app.” But the bigger opportunity is B2B. If you’ve spent 10 years selling SaaS to logistics companies, you already know the VP Ops at 20 firms moving $50M worth of freight a month. You can pitch an AI scheduling tool tomorrow and have 5 pilots running within 2 weeks. Or you’ve been an underwriter at a major insurance company. You know how slow document review drags deals out. That network means you can drop an AI summarizer straight into workflows with zero cold outreach. Or you’ve worked in hospital administration. You’ve seen firsthand how discharge paperwork bottlenecks patient flow. With the right contacts, you can pilot an AI documentation assistant in a regional hospital network on day one. The paradox is that the people with the most technical talent often have the least distribution. Meanwhile, someone with deep niche connections can get their wrapper to $1M ARR before a technical founder signs a single contract. The paradox is that the people with the most technical talent often have the least distribution. Meanwhile, someone with deep niche connections can get their wrapper to $1M ARR before a technical founder signs a single contract.

Absolutely not

You think a cursor rule will fix an insta-refactor ~ 3000 lines Get a load of this guy

This system fixes the biggest problem with vibe coding: You’re not dumping everything into the IDE and praying. You’re giving it a roadmap. You’re keeping it on rails. You stay in control. This workflow lets you ship clean, testable AI-assisted code - without the spiral. Normally I'd ask you to follow me for the playbook but this is literally it. Good luck

Let a brother eat

Literally 0

What would you do if I saw your company (YC S25) and decided to compete? If I spun up a Framer landing page, ran cold email to 1000 prospects to validate my feature set, loaded $2000 into Cursor, put on Linkin Park and built the MVP over a weekend? What would you do?

SF after your ChatGPT requests use up all the water

BREAKING 🚨: Google released Gemini 2.75

To their credit, it’s not obvious. There are accounts with thousands of followers, verified, posting what I believe to be interference and social conditioning. It’s impossible to tell.

bro said meow

if i had 7 days to build an ai agent that actually delivers value, here’s exactly what i’d do: --- please note this isn't an n8n template. you bookmarked 5000 of those posts and never built one, you might as well get some new information --- this isn’t theory. this is how i went from 0 to 5-figure mrr in 2 weeks. no funding, no team - just shipping real stuff that works. day 1: pick a real, painful problem you’re not building a demo. you’re solving work someone already pays for. best problems are high volume, repetitive, low risk if something breaks. think refunds, address changes, crm updates, lead qualifying, internal ops. if the user already has a script or sop, even better, the agent just follows instructions instead of guessing. day 2: build the skeleton skip ui. focus on the loop: receive an input (webhook, email, chat), run it through an llm, trigger a real action (refund, update, notify). use any stack you know: n8n, lambda, make, just get input to output working end to end. day 3: classify, don’t generate this is where most agents break. don’t ask “what should we do?” and risk hallucinations. instead, frame it as classification: “here’s the message + customer metadata - pick one of these allowed actions and reply with the label only.” wrap it in retry logic and catch garbage responses. if it fails, backstop with regex rules, a second model, or route to a human. goal: confidence, not creativity. day 4: map text to real actions a chatbot only replies with words. a real agent calls apis, updates orders, cancels shipments. have the llm return structured data (e.g., {"intent":"refund","order_id":"123"}), then hardcode mappings to run the logic. control everything. that’s what turns a toy into a tool. day 5: plug into their stack make adoption frictionless: connect your agent to what they already use - shopify/zendesk for ecommerce, hubspot/notion/sheets for ops, slack/email for comms. this is how you get people saying “wow, this just works.” day 6: run it in the wild get real data flowing. break things. fix them. log inputs, outputs, response times, errors. doesn’t need to be fancy - pipe logs to a google sheet or your inbox. your agent will fail sometimes. learn where and why. day 7: give it to someone deploy for a real business - ideally one with lots of grunt work and no engineering team. don’t charge. just show it works. if it saves them time or money, you’ve got a case study and proof. most people overthink ai agents. you don’t need rag or embeddings or a massive framework. solve a real problem, end to end, with something that runs. one flow, one user, one painful task, repeat until it works. everything else is noise. i write this stuff for people building ai agents that actually do work. if you want my full build kit and future breakdowns, join here: varickagents.kit.com/join. i only send something if it’s genuinely useful. cheers.

Anyone hiring software engineers? I made this at Meta and got fired.

Your vibe coded app probably uses Supabase. And if you're trusting Cursor with your DBs, you've probably broken it a few dozen times already. Do this instead: Before blindly copy and pasting, or even worse letting Cursor change your Supabase via the CLI, feed it this prompt: "If your changes involve the DB schemas in Supabase, first give me a read-only script that I can copy and paste into Supabase that will clarify any questions or assumptions you are making about the DB schemas. Use UNION to join queries together so the script shows all results in one output, instead of only the last one." Cursor will then give you a SQL script that you can safely run. Run that script in Supabase → click Export → Copy as JSON → feed back into Cursor. Now Cursor has the context it needs to make changes to the DB Schemas without breaking anything. The next change it gives you is now 99% less likely to break shit.

We live in a society

bullish on: - prediction markets - custom-built ai for back-office - ai assisted programming - anti-ai products bearish on: - ai for content/ideation - generalist ai SaaS - anthropic (sorry) - traditional VC model of funding

kinda wild that icon raised millions for what is essentially an n8n template tier 1 n8n template surely

After an incredible 3 years as a Software Engineer at Meta, I've made the decision to leave and start my own venture: Varick Agents, an AI Agency that builds robust and accurate AI Agents for enterprise. We've already delivered some incredible results for our early clients, and case studies are coming soon that showcase the cost savings and efficiency gains possible in today's market. The appetite for AI Agents in the office has been immense, but the platforms that exist today demand too much technical expertise from companies. For businesses with top-tier AI-trained engineers, this isn't a problem. But the average company lacks engineers with deep exposure to frontier AI technology. Which LLMs are best-suited for which tasks? How do you prompt AI effectively? How do you handle seamless handoffs between different models? These are all absolutely essential parts of building AI Agents that can actually save businesses time and money. That's where Varick Agents comes in. We're a team of AI experts: engineers from big tech who live and breathe the latest AI developments. We have a proven track record of building robust, enterprise-ready AI Agents that augment your employees' workflows, dramatically reducing task-completion time while integrating with your existing stack instead of forcing costly migrations. The AI revolution is here, and the businesses embracing it thoughtfully today will have a significant competitive advantage. If your business has workflows that could benefit from AI automation, I'd love to explore how we can help. Schedule a free consultation call with us below:

Reported for impersonation

it got my ass too

I’m tired all the time even though I sleep for 7-8 hours. Wake up groggy, feel like melting later in the day. Anyone have advice

bro got 3 of the same billboard in the same place

“Founder mode” and it’s hitting Tab in Cursor 500 times while scrolling Twitter

Half of vibe coding is prompting and half of prompting is knowing what you want to build and half of knowing what you want to build is knowing how to build it

F*cking love the Whop team. If I could go back in time, I would build Shopify Apps, but since I can't, I'm building on Whop. Their app store launched recently, which means I have first-movers advantage. I expect their user base to triple over the next year and 10x over the next two, which means I'm playing for exponentials. And finally, they don't lock you into their platform - meaning I can deploy my app on their site AND others. Genuinely a no-brainer, and as always incredibly bullish. Stop building for nobody and start building on Whop. Also on the low I have insider information that says this is about to get even more lucrative - stay tuned.