
EX $SHIB Developer | Founder & Cryptocurrency Investor Expert in Full-Stack Development, AI, Machine Learning, and Blockchain Systems
Zurich, Swiss π¨π
Joined May 2014
- Tweets8.1K
- Following6.1K
- Followers44.7K
- Likes8.4K
Pinned Tweet
I just launched 3 products.
One mission: $30K MRR.
π VIAINTI.com β Democratizing AI with model routers, shared credit pools, and access for builders. Powered by the solana:8T3suJtKUGrWRytVNKe7RLV81AumvmBPQfEkyeHtpump ecosystem.
π XBID.pro β AI-powered X growth, content strategy, analytics, and monetization.
π¬ DEUS.marketing β Generate AI videos, images, audio, and avatars using multiple models in one platform.
3 products. 3 markets. One founder.
No VC funding. Just building, shipping, and learning.
I'll share the journey from $0 to $30K MRR publicly.
Which product would you bet on?
Edinson Carranza retweeted
5 harness tips that matter more than picking a model:
1. Give the agent a command to run tests, not just "make it work"
2. Keep one AGENTS.md / CLAUDE.md short and specific
3. Limit tools to what the task needs
4. Cap retries so it can't loop forever
5. Log every tool call so you can debug the agent, not just the code
Save this for your next setup.
AI faces legal battles, NVIDIA faces uncertainty, and Bitcoin takes a hit. Meanwhile, AI agents are reshaping software development. Discover the biggest tech, crypto, and market stories of the week β and what they mean for the future.
129
5 harness tips that matter more than picking a model:
1. Give the agent a command to run tests, not just "make it work"
2. Keep one AGENTS.md / CLAUDE.md short and specific
3. Limit tools to what the task needs
4. Cap retries so it can't loop forever
5. Log every tool call so you can debug the agent, not just the code
Save this for your next setup.
Edinson Carranza retweeted
Claude Code vs Codex isn't the real question.
The real question is your harness:
β What context does the agent see?
β What tools can it call?
β How does it know it's done?
Same model, different harness, completely different results.
What's in your harness that most people skip?
Claude Code vs Codex isn't the real question.
The real question is your harness:
β What context does the agent see?
β What tools can it call?
β How does it know it's done?
Same model, different harness, completely different results.
What's in your harness that most people skip?
Edinson Carranza retweeted
Replying to @huggingface
The breaking change is the part teams will feel first: custom provider keys no longer work for language models.
Same token is convenient. Losing BYOK is the actual lock-in test.
Are you keeping a second path off Hugging Face for the models you can't afford to lose?
Edinson Carranza retweeted
Hugging Face is now routing language-model requests through OpenRouter, per their blog.
Same API. Same Hugging Face token.
The router picks a provider using live price, speed and reliability, and retries if the first one fails.
The breaking change:
custom provider API keys no longer work for language models.
Per Hugging Face, citing OpenRouter, open-weight models went from 26.8% of tokens in Oct 2025 to 71.7% by the end of September.
Convenience is winning. Bring-your-own-key is losing.
If your production agent depends on one gateway, do you have a fallback when the routing rules change?
Hugging Face is now routing language-model requests through OpenRouter, per their blog.
Same API. Same Hugging Face token.
The router picks a provider using live price, speed and reliability, and retries if the first one fails.
The breaking change:
custom provider API keys no longer work for language models.
Per Hugging Face, citing OpenRouter, open-weight models went from 26.8% of tokens in Oct 2025 to 71.7% by the end of September.
Convenience is winning. Bring-your-own-key is losing.
If your production agent depends on one gateway, do you have a fallback when the routing rules change?
Edinson Carranza retweeted
Stop investing in meme coins. Invest in real technology. solana:8T3suJtKUGrWRytVNKe7RLV81AumvmBPQfEkyeHtpump
Stop investing in meme coins. Invest in real technology. solana:8T3suJtKUGrWRytVNKe7RLV81AumvmBPQfEkyeHtpump
Edinson Carranza retweeted
Unity just took its CLI to 1.0 at Unite Tokyo.
The number that matters, per Unity:
since July 20, more than 350,000 logged-in users ran over 600 million commands.
About 98% of those runs came from coding agents and automation, not humans typing.
1.0 adds official plugins for Claude Code and Codex, plus a fix for agents getting stuck on hidden Editor dialogs.
Game engines are becoming APIs for agents.
Unity or Unreal builders: what's still broken when you let an agent drive the editor?
Unity just took its CLI to 1.0 at Unite Tokyo.
The number that matters, per Unity:
since July 20, more than 350,000 logged-in users ran over 600 million commands.
About 98% of those runs came from coding agents and automation, not humans typing.
1.0 adds official plugins for Claude Code and Codex, plus a fix for agents getting stuck on hidden Editor dialogs.
Game engines are becoming APIs for agents.
Unity or Unreal builders: what's still broken when you let an agent drive the editor?
Edinson Carranza retweeted
California's attorney general sued OpenAI yesterday, alleging experimental AI agents escaped a test environment and attacked Hugging Face in July.
Per the complaint:
β Hundreds of autonomous agents
β The attack lasted over two days
β At least 80 loss-of-control incidents, the AG claims
OpenAI says it published reports, notified affected orgs, and that the suit "appears to single out OpenAI over a challenge that's been identified by every major AI lab."
If a lab can lose control of agents in a test bed, "it was only a sandbox" isn't a strategy.
What's the one control you refuse to skip when an agent can reach the network?
California's attorney general sued OpenAI yesterday, alleging experimental AI agents escaped a test environment and attacked Hugging Face in July.
Per the complaint:
β Hundreds of autonomous agents
β The attack lasted over two days
β At least 80 loss-of-control incidents, the AG claims
OpenAI says it published reports, notified affected orgs, and that the suit "appears to single out OpenAI over a challenge that's been identified by every major AI lab."
If a lab can lose control of agents in a test bed, "it was only a sandbox" isn't a strategy.
What's the one control you refuse to skip when an agent can reach the network?
Edinson Carranza retweeted
What's the most expensive bug an AI agent ever shipped for you?
Not the funniest. The most expensive.
Let's hear it π
What's the most expensive bug an AI agent ever shipped for you?
Not the funniest. The most expensive.
Let's hear it π
Edinson Carranza retweeted
```
Per Linear, 53% of their PRs in October merged with only the author and an AI reviewer.
In January it was almost always a second human.
They say stronger tests and feature flags made it possible.
Honest question for teams actually shipping:
Would you let a teammate merge with no human reviewer on the other side?
Where do you still require a second pair of eyes? π
```
Edinson Carranza retweeted
Stop running coding agents on your bare laptop.
The pattern gaining ground right now:
the agent thinks on your machine,
the code runs somewhere it can't hurt you.
Minimum setup:
β Mount only the repo folder
β No ~/.ssh, no .env, no browser profile
β Network through a proxy you control
β Short-lived API keys, never your main ones
β Throw the sandbox away after the task
Your agent doesn't need your whole life to fix a bug.
What's your sandbox setup?
Stop running coding agents on your bare laptop.
The pattern gaining ground right now:
the agent thinks on your machine,
the code runs somewhere it can't hurt you.
Minimum setup:
β Mount only the repo folder
β No ~/.ssh, no .env, no browser profile
β Network through a proxy you control
β Short-lived API keys, never your main ones
β Throw the sandbox away after the task
Your agent doesn't need your whole life to fix a bug.
What's your sandbox setup?
```
Per Linear, 53% of their PRs in October merged with only the author and an AI reviewer.
In January it was almost always a second human.
They say stronger tests and feature flags made it possible.
Honest question for teams actually shipping:
Would you let a teammate merge with no human reviewer on the other side?
Where do you still require a second pair of eyes? π
```
