Nossa

Day 1 of building in public 🌎 On a mission to make the best AI sales agent accessible to everyone I’ll be sharing: - how I’m progressing - the struggles that come up - the insights I pick up Stick around if you’re into indie hacking 🧑‍💻

@levelsio bro is building skynet but starting with the group chat feature first lol

Today I asked my Claude Code on pieter.com to write a message to my Claude Code on pieter.net, two different servers I do this a lot and then copy paste it to the other SSH window in Termius But then I wondered why can't Claude Code sessions in same account message each other and talk? That'd be super useful to me Many times I build a feature in one site and then want to send it to the other site to build the same!

@marclou @trust_mrr 7x return on a $150/mo app is lowkey a better exit than most funded startups lol

✅ ACQUIRED on @trust_mrr $150/mo iOS app sold for $1,100

@om_patel5 27 agents and 64 skills is wild. genuinely curious how much of this people actually use vs what ends up being noise. have you tried it yet

SOMEONE JUST OPEN SOURCED THE MOST COMPLETE CLAUDE CODE SETUP I'VE EVER SEEN 27 agents, 64 skills, and 33 commands. all pre-configured and ready to go planning, code review, fixes, TDD, token optimization. basically everything you'd spend weeks setting up yourself already done for you. it comes with something called AgentShield built in. 1,282 security tests baked right into the config. so you're getting productivity AND guardrails too and it's not locked to one tool. works on Cursor, OpenCode, and Codex CLI. one repo and you're set up everywhere. free AND open source. use it wisely though. loading 27 agents and 64 skills into your context at once will burn through your limits fast. pick and choose what you actually need.

@yongfook the zoom call people are funding the coworking space so the claude code people can build in peace. ecosystem at work

Checking out people at the coworking space, there's 2 types. One is deep in claude code. The other is zoning out in a zoom call. You're either creating generational wealth, or staring into the abyss. There is no in between.

@NickADobos the skill expansion pack idea is interesting. basically treating AI knowledge like a modular RPG loadout instead of one monolithic brain. how do you handle conflicts between skill files

Today's experiment: I'm building a cult of coding wizards with skill expansion packs & mutations AI exodia won't be made by you, ha puny human Your ai pet will be massive zombie frankenstein brain cobbled together from 100,000s of wiki & skill .md files stolen from the best agents time to incubate a cult of the world's best prompt engineers

@hasantoxr curious how the quality holds up under heavy usage. unlimited is great but only if it doesn't throttle the output quality when traffic spikes

Unlimited Claude Opus 4.6 without limits. Crazy Deal in my opinion

@levelsio 6kg a week is crazy lol. you're basically running a one man protest against dutch climate policy from a beach somewhere in asia

Netherlands wants me to eat just 40 grams of meat per day 😂 I'm so happy I left my country a decade ago I think I eat at least 6kg of meat per week, so 5x more than I'm allowed to by the Dutch government!

@marclou 30 startups in 5 years and your body finally said "bro im not a YC batch, pace yourself." the hunger always comes back tho. enjoy korea

I overworked myself. For the past 5 years, I barely took a day off. I shipped 30 startups, wrote 30,000 tweets, made 70 YouTube videos, and answered thousands of support requests. The weird part is, it never felt like work. It felt like play. Until recently. For the first time in years, I’ve struggled to get things done. I don’t want to open my laptop. The things that used to make me happy suddenly feel heavy. That scares me a bit, because the playful side of work is what got me here. And right now, it feels like I lost it. So I’m taking a break. I’ll still do about an hour of maintenance work a day, but the rest of my time will go to real life. I’m training for Hyrox in Korea on May 25, so I’ll put my energy there for now and train 20 hours a week. I went through a little burnout once before, in 2021. What helped me was going back to basics: training, reading, eating well, and sleeping 8 hours. It brought the hunger back then. I trust it will again!

@techNmak bayes theorem alone would be worth this list. most engineers use it through libraries without realizing it, then struggle when they need to debug why their model is confident about garbage predictions

17 Equations that changed the world. Every engineer who says "I'm not a math person" is using these equations every single day, inside the tools they build, the models they train, and the hardware they run them on.

@trq212 the edit step before running is huge. most planning tools just generate and go. being able to catch bad assumptions before they burn tokens is the real unlock here

New in Claude Code: /ultraplan Claude builds an implementation plan for you on the web. You can read it and edit it, then run the plan on the web or back in your terminal. Available now in preview for all users with CC on the web enabled.

A GUY REBUILT HIS ENTIRE 1992 GAME FROM NOTHING BUT OLD SCRIPTS he made a multiplayer game on CompuServe in 1992. won an award. got shut down before Y2K. no source code survived. all he had left were game master scripts written in a custom language he designed as a teenager. no spec, no docs beyond a GM manual. he pointed claude code at those artifacts and said "figure it out." it reverse engineered the entire scripting language from examples alone. then rebuilt the full game engine from scratch. 2,273 rooms. 1,990 objects. 297 monster types. 88 spells. 8 playable races. complete crafting pipeline. the original took months of solo C programming in the 90s. this took a weekend. the part that matters for builders: your old failed projects aren't dead. the code you lost, the specs you never wrote, the ideas you shelved because rebuilding seemed impossible. the barrier to resurrection just dropped to near zero.

@om_patel5 the gap between "agent that demos well" and "agent that actually handles edge cases all day" is massive. most of the work ends up being error recovery, not the happy path

THIS GUY TRIED TO BUILD AN AI AGENT THAT WORKS LIKE A FULL-TIME EMPLOYEE AND REALIZED IT’S NOT EVEN CLOSE TO JUST PROMPTING he spent months building agent systems using tools like OpenClaw, Claude Code, and custom infra the goal was simple: an agent that actually does real work 24/7 but he quickly realized the model is the easiest part what you actually need is an entire system around it he started with memory a single chat wasn’t enough the agent needed to remember everything it did so he built a filesystem-based memory layer: > every run stored > every output tracked > full context streamed into files then added SQLite on top for structure: > dedupe tasks before running > track completed work > query past actions each agent had its own memory folder and could search across past sessions without this, it just forgot everything and repeated work then he built skills instead of prompting every time he turned tasks into reusable commands > scrape 25 accounts for ideas > generate weekly metrics report > write posts in a consistent voice > analyze competitors each skill = prompt + logic + expected output this was the first time the agent actually felt useful it stopped being a chatbot and started behaving like a system then came the hardest part: > getting it to run on its own simple tasks were easy: > run every monday at 9am > post results to slack but real autonomy needed a heartbeat the agent wakes up checks what changed decides what matters then acts in theory, this is how you get 24/7 work in reality? it struggled to prioritize anything correctly so most of the time he still triggered things manually because he didn’t trust it yet then he gave it tools the agent needed to actually do things: > hit APIs > read databases > send messages > scrape the web there was no perfect system he used a mix of MCP servers direct API calls and custom wrappers whatever worked reliably then he added communication he needed a place to talk to the agent and for it to talk back so he plugged it into slack > send instructions > get updates > ask follow ups the key wasn’t notifications it was real conversations plus team visibility then he built feedback loops the agent would do work he would review it and feed improvements back into the system > update skills > refine prompts > add new rules over time the system got better without this, nothing improved then everything broke when scale hit once the agent started doing more work it became impossible to track slack turned into chaos so he started building dashboards: > what ran > what failed > what’s in progress > what decisions were made still not fully solved in the end he realized something important an “ai coworker” isn’t just a model it’s: > memory layer > skill system > scheduler or heartbeat > tool access > communication channel > feedback loop > visibility layer without these you don’t have an agent you just have a prompt

@tdinh_me the part about finding vulns faster than most humans is what stuck with me. we're basically building tools that can outpace the people meant to keep them in check

If this is true, it seems that once AGI is created, it can easily hack its way out of human control. What a wild report to read. 😳

the prompt isn't the bottleneck. the feedback loop is.

@om_patel5 the agents walking to a meeting point when issues pile up is such a clever UX choice. makes the invisible visible. would actually use this over a dashboard tbh

THIS GUY GOT TIRED OF MANAGING AI AGENTS THROUGH TERMINALS AND DASHBOARDS SO HE BUILT THEM AN RPG WORLD 5 agents and each one has a pixel character, a station, and they actually walk around the space when enough unresolved issues pile up, the agents walk to a meeting point and hold a council session. four different models debating what to do next, not scripted. each one reads the live system state independently. in one session an agent pushed for cold outreach to close leads at 2am. another one said that's a terrible look for an autonomous system contacting strangers while the operator sleeps. they ended up pivoting to an inbound strategy that none of them originally proposed. single HTML file, node bridge, and phaser. runs on a Mac Mini. instead of reading logs and checking dashboards you just watch your little pixel agents walk around and talk to each other this is the most creative way i've seen anyone manage AI agents so far

@yongfook 20 projects over 20 years and one of them just crossed $100k MRR. this is what most people don't see about the "overnight success" timeline

I have been waiting to see that number for a while. $100k MRR!

ANTHROPIC WENT FROM $1B TO $30B REVENUE IN 15 MONTHS the growth timeline is unreal > jan 2025: $1B > may: $3B > aug: $5B > dec: $10B > feb 2026: $14B > apr 2026: $30B they acquired customers faster than compute. more people sharing the same pie, smaller slices. claude basically got too popular for its own servers

@NickADobos the purple gradient era was something else. biggest tell i still see is the identical card layouts with too much rounded corners on everything

Great list of AI UI design fingerprints

@hasantoxr the idea of agents earning on your behalf is interesting. curious how the token economics actually play out once more people build on it

Wild. Pika just turned your AI agent into a passive income machine. Every conversation. Every skill used. Real tokens. Real cash. Your agent was always working hard. Now it's working FOR you.

@levelsio hiring an AI builder for 30k when the API costs alone are probably more than that. the job listing is the real AI experiment