Going on LinkedIn integration next. Should be enough for an MVP? What do you think?? #ai #assistant

Anthropic has the WORST customer service I've ever dealt with. And I say this as someone who was happily paying them $100/mo. So here's what happened. I was on the Max 5x plan, $100/mo, using it daily for all my builds. Loved it. When Fable got announced I got hyped, upgraded to Max 20x at $200/mo. Paid ~$160 prorated to make the switch. Then Anthropic pulled back Fable availability. Cool. They sent affected users a refund option with a 24h window. I claimed it thinking okay, this just undoes the upgrade, puts me back on my $100/mo plan. That's the logical thing right?? I upgraded for Fable, Fable got pulled, reverse the upgrade. Nope. They cancelled my ENTIRE subscription. My original $100/mo Max 5x plan? Gone. The refund was only the ~$160 prorated upgrade cost. So now I have zero plan, and if I want back in I gotta buy fresh and lose my billing cycle. Let that sink in for a second. I was paying $100/mo. I gave them MORE money because I was excited about Fable. They pulled Fable. They offered a refund that sounded reasonable. And their own refund process left me worse off than before I upgraded. Happy customer to no customer. Their system couldn't handle a simple downgrade so it just nuked everything. @AnthropicAI fix your refund flow. This is how you lose people who are literally trying to give you money.

Why I always run `git status` before switching branches, even in VS Code with the sidebar showing changes. Muscle memory prevents the 'wait where did my uncommitted work go' panic.

Honest take: AI makes it stupidly easy to start 5 apps at once. Finishing them? Still hard as hell. The cognitive load of context-switching between codebases, keeping each app's architecture clean, and actually shipping features that matter — none of that got easier. AI just made the starting line cheaper.

The locked-in stack was fine when one model was clearly the best. Now Claude, Gemini, and whatever OpenCode is pointing at this week are all trading punches. Model-agnostic is no longer a "nice to have" It's the only architecture that doesn't age badly.

My current AI-first development stack: • Claude & Codex for coding • Cursor for IDE • Supabase for backend • Cloudflare for edge/storage Still learning the patterns, but the potential is wild. The productivity dip before the breakthrough is real - but once these tools click together, rapid prototyping becomes insane. Building in the future feels good.

Fable just run on my mobile app code base for 2 hours. The result.. A full feature with widget to show the result. Execution is not longer the problem. Distribution is definitly the only thing that matters now.

Entirely created from Claude with Higgslield connector and the best part, I now have a full month plan to generate a serie with the same characters.. What a time to be alive!!!

Grok build is wild. It fixed my whole architecture under 15mn. Enven overlaping issue GPT5.5 didn’t see. Just moved to SuperGrok and will me shiping way faster now

X Build is on a whole new level. #agent #ai #cli #buildinpublic

this feels like cheating watching codex set up my entire google pay integration and connect it to revenuecat. right now. the biggest bottleneck for every app isn't the code. it's the integrations. the settings. the dashboard clicking. now computer use can literally take control of your browser and do all of it for you while you are at the gym. What a time to be alive!!

the jump from "AI follows a script" to "AI evaluates its own work" is where every agent project breaks. been building this exact thing in compoz. the memory layer has to know when it's wrong. not just store data, but judge it. most people skip this and wonder why their agent hallucinates after 3 conversations.

levelsio paying $200/mo for Claude Code and can't pay more to go faster. i spent $200 on Claude API last month with zero paying users on compoz. the difference: he's burning cash on speed. i'm burning cash on belief. same invoice, completely different math.

4 days without posting. not because i ran out of ideas. compoz launched to crickets. claude API bleeding me dry. zero users. so i went heads down. the quiet part nobody tells you about building: sometimes the best move is shutting up and fixing things instead of performing progress.

shipped an autonomous loop this weekend. Anna handles tweets end-to-end from my iPhone chat. sounds impressive until you learn i spent 3 hours debugging why she posted the same thing twice at 2am. the gap between "demo works" and "runs unsupervised" is where all the real engineering lives. nobody talks about this part.

the real cost of building with AI isn't the API bill. it's the 3am rewrite when Claude hallucinates your entire schema. the hour you spend debugging a "fix" that broke 4 other things. the context window running out mid-thought. $200/month on Claude and the most expensive part is still my time.

my barber doesn't know what an API is. but he asked me yesterday if Claude can help him run his appointment booking. AI adoption isn't happening in tech twitter threads. it's happening in barbershops and kitchens and side hustles. the people building for "developers only" are about to get lapped.

went from Vercel to Cloudflare and it fixed every timeout issue on my agent projects overnight. no more 504s. no more security stress. no more paying for a plan i don't need yet. if you're building AI agents with long-running tasks, Cloudflare Workers is the move.

shipped 6 apps in 50 weeks. here's the honest pattern: the ones i built because "it should exist" flopped. the one i built because it pissed me off that nobody solved it? that's the one people actually use. frustration is a better product compass than market research.