CTO @ZKcandyHQ Founder @ceo__ai author of ERC-7662 draft standard for AI Agent NFTs

Joined November 2011
Last week I went from a napkin idea to a live, deployed website for a business that connects Latin American workers to UAE employers. The entire thing - strategy, brand, ICPs, domain, website - was built through AI. Here's exactly how it happened. (a few thoughts)
1
1
253
The warnings going around right now about AI capabilities should be heeded. There are a couple of immediate solutions that would do a lot to prevent the worst from happening: (1) The major cloud providers like AWS, GCP, Alibaba Cloud etc. should publish what they are doing to prevent escape. Anyone capable of running AI at cloud scale.
I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.
1
1
95
(2) Anyone working on self-improving superintelligence should have to open-source their work. Let us all see the risks you are taking so we can all be part of finding the mitigation solutions. I think this last one is what @elonmusk had in mind with OpenAI in the first place.
1
53
Unfortunately a lot of the most exciting new capabilities of AI are also its most dangerous. The likelihood that someone will cut corners under time / money / market pressure is too high to ignore.
18
What if the current AI slop markers of today become like a hipster cool thing to put on T-shirts 20 years from now. Could happen. Maybe even sooner than that.
53
Most 'side project failure' isn't founder failure. It's orphaned-software failure - the build never carried the intent, so maintaining it felt like pushing a stranger's cart. Fix the translation layer and maintenance becomes a river.
65
You don't have a productivity problem. You have an idea-hoarding problem.
1
108
Proof, boring numbers: TrabajoEmiratos.com: napkin to deployed full-stack business in ~16 minutes. 102 files. ~65,000 lines. Live Spanish site an hour later. Not a demo - a portfolio asset. The code carries the strategy.
1
69
New decision tree: Every executed idea gets a verdict: keep, expand, discontinue, sell. Status changes weekly. Nothing sits at 'working on it' for months. Validation comes after a high-fidelity launch and real marketing - not 40 clicks on a landing page.
47
The translation layer is the cost. Agencies: $150–400K, 6 months, 12 handoffs - software arrives drifted. Prompt-to-app builders: 20 minutes - orphaned on arrival. The problem was never speed. Strategy and execution must be one continuous act. #VentureBuilding
1
156
Your last build - idea to ship. Count the handoffs.
96
Building software has gotten very cheap. And very orphaned. The maintenance collapse is coming. Here's why:
1
157
Proof this isn't a theory: TrabajoEmiratos.com. From a napkin idea to a deployed, 102-file, 65,000-line, self-documenting full-stack business. 16-minute app build. Live Spanish 30-page website in 1 hour. The code comments explain *why*, not just *what*.
1
84
The next decade belongs to Builder who can maintain what they create. Know your next move. Then watch it get built. Which side of the pile are you on? #BuildInPublic #VentureBuilding
74