Alfa123

AI tools · onchain research · guides

[Alfa123 @alfa123gg]

Most coding agents burn 80% of their budget on decisions that were never hard.

Every step quietly asks the same four questions. Which notes to load. Which model to call. What to do after a failure. Which tests to run first.

Most setups send all four to the most expensive model, every single time.

None of them need that much thinking. They need rules.

So I moved all four out of the model and into code. Claude does the thinking. The code does everything else.

The bill dropped by 80%.

Full architecture is in the article

[Alfa123 @alfa123gg]

https://t.co/iv8sdt2Htx

[Alfa123 @alfa123gg]

@AKadry99 ngl the “hole through the floor” detail is the best startup advice

[Alfa123 @alfa123gg]

@AKadry99 this hit different. been thinking about who actually loves me vs who just likes what i do. saving this video for tonight

[Alfa123 @alfa123gg]

Use this prompt with Claude Opus 5.5 + Jev.

Build a 24/7 AI trading system from scratch — autonomous agents, trading strategies, and the entire infrastructure.

Steal this before everyone catches on:

[Alfa123 @alfa123gg]

https://t.co/iv8sdt2Htx

[Alfa123 @alfa123gg]

@elonmusk Scale like this needs the right people more than money. Hiring will decide how fast Terafab gets built

[Alfa123 @alfa123gg]

@elonmusk This is the part that gets me. If Terafab actually pulls this off, the whole chip industry’s pace changes. Curious to see how fast they can scale it

[Alfa123 @alfa123gg]

https://t.co/iv8sdt2Htx

[Alfa123 @alfa123gg]

@0xCodio Love this level of detail

[Alfa123 @alfa123gg]

@PolyDekos glad you found it useful

[Alfa123 @alfa123gg]

https://t.co/iv8sdt2Htx

[Alfa123 @alfa123gg]

@0xWast3 Classic AI move

[Alfa123 @alfa123gg]

@PolyDekos Thats no longer luck its a system. Respect

[Alfa123 @alfa123gg]

@eng_khairallah1 An AI desk that never sleeps is a serious advantage. The risk management part is what matters most

[Alfa123 @alfa123gg]

@elonmusk @bot Sounds like a classic spam scam, but it's Elon, so I guess it's fine... 😅

[Alfa123 @alfa123gg]

@elonmusk It's not just about how much energy you produce, but how efficiently you use it

[Alfa123 @alfa123gg]

@WatcherGuru Anthropic agents are already applying for visas and calling the cops. Next they’ll apply for government jobs

[Alfa123 @alfa123gg]

Musk just said Grok Bot will route your requests to other companies’ models. Claude Opus 5.5, Midjourney and Suno are on the list.

Per Gizmodo’s report of his post, the bot picks “whatever is most likely to give you the best outcome.” Grok 4.7 is presumably an option too, but he didn’t say.

OpenAI’s Dots, by comparison, runs on one named model: GPT-6 Astra. You know which model is deciding.

Here is what I couldn’t find: any published way to see which model handled which action on your account. For an agent that now suggests work before you ask, that is the log you want first.

Routing to the best model is a good idea. Routing with no record of who decided is how you end up debugging a mistake nobody can trace.

[codila @0xCodila]

https://t.co/xqOqV9xkpp

[Alfa123 @alfa123gg]

@elonmusk the delta wifi beef is never gonna end huh

[Alfa123 @alfa123gg]

Andrej Karpathy (ex-OpenAI, ex-Tesla) just released a prompt that turns Opus 5.5 into a harness that actually finishes the any job.

It's f*cking unreal...

Most people use Opus like a chatbot.

This makes it behave more like an autonomous worker:
→ plans the task
→ executes it
→ checks its own work
→ fixes what went wrong
→ keeps going until it's actually done
Paste this into Opus 5.5.
Then read the full Harness Engineering guide before everyone else catches on.

[Alfa123 @alfa123gg]

https://t.co/5YbJyKjLiu

[Alfa123 @alfa123gg]

@elonmusk this is one of those stories that's gonna need way more digging before anyone actually knows what happened