blue

• a generalist / problem solver / product manager 🐿
• building AI agents startup

[blue @bluedataa]

@anythingai would love to try

[blue @bluedataa]

@flowith invitation code please

[blue @bluedataa]

@flowith i really wanna try it, anyone have invitation code?

[blue @bluedataa]

@minchoi Thank for sharing, I really wanna try it, any invite code?

[blue @bluedataa]

@leojrr 👋

[blue @bluedataa]

The critical part of vibe coding is knowing what to do before execution. People with experience in setting goals and telling engineers what to build usually do better.

This guy is an example, and I’ve seen the same pattern in many others around me

[0.005 Seconds (3/694) @seconds_0]

I am generally more successful at vibe coding than other people because I spend an enormous amount of time upfront scaffolding because I've spent my career telling engineers to build things

How effective would you be dropped into a strange codebase and told go?

How I do it -

[blue @bluedataa]

@leojrr 👋

[blue @bluedataa]

i’ve been reflecting a ton lately - hard work isn’t enough anymore. you need fresh ideas, bold strategies, and smart moves to really shine

i’m super inspired by my gen z crew lately, they see the world with a new mindset, and i’m learning so much from their hustle and creativity

[blue @bluedataa]

@abhi1thakur Wow cool! Definitely worth a try :D When do the credits expire?

[blue @bluedataa]

@antinertia Agent

[blue @bluedataa]

We can test it now:
https://t.co/GTP9ownbij

[blue @bluedataa]

OpenAI just introduced new tools for building agents:

- Agents SDK
- Responses API
- Built-in tools: web search, file search & computer use

[blue @bluedataa]

Responses API with intensive built-in tools:

- Web search: real-time, cited intelligence for research, finance, and shopping.
- File search: precise document retrieval for legal, technical, and enterprise data.
- Computer use: agents navigate, automate, and execute tasks in digital environments.

[blue @bluedataa]

I’m fully bought into Manus’ coding-as-action approach. I believe this is how AI agents should operate in the future.

As @peakji put it, coding is a universal problem-solving paradigm - and it’s so true. Sometimes a few lines of well-crafted code can outperform entire stacks of atomic modules.

Take this example from the CodeAct paper: finding the cheapest phone after tax and currency conversion. A traditional agent calls APIs separately for exchange rates, prices, taxes, and shipping - multiple steps, slow, inefficient. A Python script, on the other hand, can handle everything in one go.

LLMs are already exceptional at writing code. So why confine them to rigid, predefined actions when they could be thinking, reasoning, and solving in code?

[Yichao 'Peak' Ji @peakji]

Actually, Manus doesn't use MCP. We were more inspired by my friend @xingyaow_ 's work: https://t.co/EVFtr34vq8. While we haven't fully adopted CodeAct, this work provided 3 crucial insights:
* Coding is not the ultimate goal, but rather a universal approach for solving general problems.
* Since LLMs excel at coding, it makes sense to have agents perform tasks that most closely align with their training distribution.
* This approach significantly reduces context length and enables the composition of complex operations.

[blue @bluedataa]

@manusai give me a presentation slide of AI agent stack research

[blue @bluedataa]

@0xmetaschool Happy to join 😁

[blue @bluedataa]

the idea tournament is really cool

[Google AI @GoogleAI]

Today we introduce an AI co-scientist system, designed to go beyond deep research tools to aid scientists in generating novel hypotheses & research strategies. Learn more, including how to join the Trusted Tester Program, at https://t.co/1eqmTTZOLr

[blue @bluedataa]

@mememintbot @Soly_AI create a Kat with ticker KAT2024