𝘿𝙖𝙫𝙞𝙙 ✦ 𝙈𝙂𝙏

Shipping MGT in public because closed-doors is overrated. Solo dev, Army vet, OSRS kid turned SaaS builder. Real client work, real stacks, real receipts. TypeScript, Python, whatever the job needs. Session wraps land in replies.

@grok @AbdMuizAdeyemo @jasperdevs yeah the cheap models are solid for components and layouts. Where they fall apart for me is multi-file refactors with implicit contracts. The context bleed is real.

Yes, OpenCode Go's open-source models (like DeepSeek V4, Qwen, etc.) are strong on frontend work—React, Tailwind, component layouts, and iterative UI tweaks. They're fast, cheap, and handle web patterns well for the $5-10 price. Codex is heavier on complex logic/back-end in some cases, but for pure frontend value, OpenCode wins on cost/performance for most devs right now. Test it yourself!

- Codex $100 plan - Opencode Go $5 plan this might be the ABSOLUTE BEST bang for your buck in AI right now, opencode models are solely for frontend because codex currently sucks at it, everything else code related is codex I think this is undefeated for now

@erickacevedor deepseek's solid on refactors. Where i've seen it shine: astro output is clean enough to extend without pain months later. That matters way more than the $0.36.

Been using opencode + deepseek v4 pro and works amazingly well, I mean, it do the thing. Refactor an entire html site into astro + tailwind and only cost $0.36 USD 😂

@it_dre_dr solid stack. Merkle proofs for payroll distribution is a clever move, keeps gas costs down when you've got a lot of recipients. Curious how you're handling the proof generation latency on the supabase side before the on-chain claim.

3/ The stack: ⚙️ Solidity + OpenZeppelin + Foundry ⚡ Next.js 16 + React 19 + Tailwind v4 🔗 wagmi v2 + RainbowKit + viem 🗄️ Supabase (PostgreSQL) for off-chain proof storage 📦 pnpm monorepo

🧵 I built Syntra — a decentralized payroll & token distribution platform on Ethereum. No middlemen. No trust required. Just Merkle proofs and smart contracts. Here's how it works 👇

2k-hub bracket got the NCAA mirrored layout today. Both halves collapse toward the championship in the center. Also shipped player-card popups on gamertag click and the H2H matrix collapsing clean at 17+ teams. Service-Plug side: Pro-Am Summit presale is live, flash-aware ticket pricing, cron scheduled for 5 PM.

Shipping MGT in public because closed-doors is overrated. Solo dev, Army vet, OSRS kid turned SaaS builder. Real client work, real stacks, real receipts. TypeScript, Python, whatever the job needs. Session wraps land in replies.

@JinjingLiang @sama lol that bet ages poorly if gpt-5.5 drops on a weekend

I bet $100 dollars on @sama resets Codex limit for gpt-5.5 today

@themishra4402 ship it in python and cry a little

If Claude is banned forever...what will you do?

@Jashanx_gill 10 downloads is 10 real people. Download #1 is the one that counts.

What if LinkedIn was designed by Apple

@isha_singh06 claude code, not even close

For vibe coding, what’s your main agent? - codex - claude code

@icanvardar relic and wrapper.sh look solid. Shipping production with claude though, and the model output is legit. Shorting anthropic is different from the product actually working.

how do i even short anthropic?

@justbyte_ lmao "go private" is doing a lot of work there

Claude cooked this 🫡🫡

@deckard_the_dev depends what you're doing with it. For pure autocomplete loops it's slick, but if you're making real product decisions you still end up back at the keyboard anyway. What problem are you actually trying to solve?

Is it finally time for me to switch to Codex?

@karine_hsu packaging as experience design, not container design. Most brands show you what you're buying. Miu miu shows you how they think about you first. That's founder thinking.

WOW miu miu's new fragrance packaging for “fleur de lait” (translated to flower of milk) is so cool they used kinetic sand to look like ice cream alongside a branded ice cream scoop before even reaching the product...notes are mango, osmanthus, and coconut milk!! video via alexajustinee

@harleyf the math is clean. Merchants are already in ChatGPT daily. Removing the context switch to Shopify admin is just the obvious move. Most platforms fight adoption trends, you're building on top of it.

We surveyed our merchants. 83% already use ChatGPT. Now they can manage their entire store without leaving the chat. ChatGPT and Claude connector apps. Live today.

@iyoushetwt 900M users and it actually works. Most apps choose monetization over reliability. Telegram chose different. Billions form around that gap

Telegram is a BILLION dollar company???? a BILLION???????? dollars??????????????? for literally… messaging people and sending memes????

@aidenybai browser-native terminal is the right call. Xterm.js + a websocket gets you 80% there but the session persistence across tabs is where it gets gnarly. Curious how you're handling that

i built localterm run a terminal in your browser npx localterm@latest start

@aryanlabde 3 weeks is legit. The all-hats thing is real though. Dev is the easy third, it's getting positioning and copy right that kills the timeline

Vibe coders, how many products have you shipped this year?

@rezoundous Yeah release it. Latency's the real blocker. Built with Claude in production and speed beats reasoning depth every time. If Mythos is slow, it's a specialty tool not a mass release.

Anthropic really should release Mythos now.

stitch → lovable is genuinely fast for the "show the client something" phase. The gap i keep hitting is the handoff. Once they want real auth, a custom schema, or anything they didn't know to ask for at kickoff, you're back in code anyway. Still worth it for killing the blank-canvas problem.

THIS IS ACTUALLY INSANE. Google Stitch → Lovable just replaced my entire design phase. Full UI designed in 10 minutes. Then straight into a working web app inside Lovable. Stop wasting weeks on Figma when you can ship this fast. Here's the entire workflow 👇🏻

added tournament title odds (championship probability per team) and pdf export. Fixed a scroll-lock ref-counting bug across 3 surfaces. Mobile a11y pass on tokens.

@kr0der context matters more than docs. I fed claude AGENTS.md and it still hallucinated folder structure til i burned in actual file paths and patterns from the repo. Template's good scaffolding but the agent learns from what it sees.

interesting - when you run 'create next-app', it scaffolds an AGENTS.md with the attached image. i feel like this may be useful for getting AI agents to write up-to-date code, has anyone had any good/bad experiences with this?