Bear

Building @KonagiVT 🦊
Autonomous Bilingual AI Kitsune VTuber
Rust+ML AI
Yorishiro(依り代) AI Agent Research System

Also my Indie game: @AstralRivals

[Bear @BearTheDev]

@sattyyouneed Normal people are not going to be able to optimize ai workflow as good as a dedicated engineer

[Bear @BearTheDev]

@yashmp2004 It is a skill issue tbh

[Bear @BearTheDev]

Someone asked me what code is.

I told them:

“It’s where I write a todo list, and then the electric doggy follows the commands I wrote.” 🐻💻🐶⚡

…Honestly, that’s basically it. 🐾

#Programming #Coding #SoftwareEngineering #AI

[Bear @BearTheDev]

@vivoplt Api limit

[Bear @BearTheDev]

@jiya_3063 I recently found some benefit in at least writing the core functionality by hand. The ai can then just build all the other stuff quickly for you.

[Bear @BearTheDev]

@Abhijit_091 Spend the rest of the week learning to code and minimize token consumption.

[Bear @BearTheDev]

@Officially_Dev_ Both. You need laptop to be portable and desktop to run a big GPU.

Having the laptop be a Mac has given me a lot more options, but I also have a windows laptop woth a good CPU.

[Bear @BearTheDev]

By running Claude through a local LLM, I can have my local machine do simple tasks with Claude as an orchestrator. This moves a lot of the mundane token churns on to my hardware, leading to me using less Claude api tokens.

[Bear @BearTheDev]

Ive gone full circle with vibe coding.

I am so good it at now that I hit the weekly token cap way too fast and now I have to code manually again because I dont want to put out more money for api tokens.

Though I am working on a system that allows Claude to run a local LLM.

[Bear @BearTheDev]

@the_codewala Unfortunately, yes.

[Bear @BearTheDev]

@trikcode It isnt unmaintainable if you train an ai to manage it as you make it.

[Bear @BearTheDev]

@jiya_3063 I see no reason to vibe code a new project with Java. Shipping with Flutter has been so much easier for me.

[Bear @BearTheDev]

@ebisu_ya144 I need 😫

[Bear @BearTheDev]

@VtuberSolace Strongest bear, eh?

[Bear @BearTheDev]

An idea that a fan of mine pitched to me.

We have the technology.

Yorishiro autonomous AI system is able to support multiple ai profiles with their own memory and optional settings.

The limit is hardware as so far it is
taking a 10gb vram gpu to run a low end version.

[Bear @BearTheDev]

Status: designed, building. Decisions written, queue cut, no crate yet.

v1 still runs her every day. It's the evidence, not the punchline — a wish store sitting at 23 open is what taught me that a system for wanting things needs an exit.

[Bear @BearTheDev]

Konagi v1 could want things and never finish them. Her wish store holds 23 wishes — all 23 status open.

v2 rebuilds wanting as karma: three states, taken from the Yoga Sutras. Her behavior is decided by whichever ones are active right now.

[Bear @BearTheDev]

Then the loop closes: spent karma becomes training data.

Follow-through stops being an instruction in a prompt and becomes something she learned — because every intention she kept, dropped or refused is on the record the next adapter trains on.