ToolkitSoft®

Building deterministic systems that oversee non-deterministic LLMs. Systemorama, Promptorama, Progorama for controllable AI engineering. Repeatable by design.

[ToolkitSoft® @ToolkitSoft]

@alexandriem7 @XCodeWraith Spot on. "Toss pot"... Such a great phrase. I like the word "fuckology".

[ToolkitSoft® @ToolkitSoft]

@fcuk112 I use a varier chair like this. As it is high like a bar stool you can't slouch. It's very comfortable, I don't feel I'm suffering and my posture in it is much better. I also stretch every morning for about 2 minutes which also helps.

[ToolkitSoft® @ToolkitSoft]

@lovabler0gue @nateberkopec Made me laugh.

[ToolkitSoft® @ToolkitSoft]

@simantaparidaux @paraschopra Yeah, we're really fucking acing it at the moment.

[ToolkitSoft® @ToolkitSoft]

@sama It's not messy. Terrific product. Decent treatment of users.

[ToolkitSoft® @ToolkitSoft]

@innerexcellence You paying through Bruxelles by any chance?

[ToolkitSoft® @ToolkitSoft]

@DanielPink Good post. I always try to remember the maxim "fear is wasted creativity".

[ToolkitSoft® @ToolkitSoft]

@kuizinas Wondering the same thing....

[ToolkitSoft® @ToolkitSoft]

@richardhutton I run on Ultra all the time. Heavy loads. Have a spec driven approach. Don't spend money on usage credits. Create a second account.

[ToolkitSoft® @ToolkitSoft]

@samueljmcd You can architect against that. In my framework all output from the LLM is an XSD validated XML format. A camel through the eye of a needle architecture. Solves the problem and the LLM has to shut the fuck up basically. €€€ saved.

[ToolkitSoft® @ToolkitSoft]

@doodlestein I always use Ultra and don't have this issue. I use a spec driven approach and a custom harness doing as much of the heavy lifting as possible. (Zero Token Architecture philosophy)

[ToolkitSoft® @ToolkitSoft]

@PremiumGoblin On one level it's impressive. However say you had a new game idea, not so sure an AI would do such a good job due to a lack of prior art.

[ToolkitSoft® @ToolkitSoft]

@GregSkerman @JackEllis Indeed.

[ToolkitSoft® @ToolkitSoft]

@JackEllis Sorry @JackEllis, get with the program. 2026 it's all about the trillions.

[ToolkitSoft® @ToolkitSoft]

I'm working on a "landscape" module where every resource is represented in a graph (I need to cover Azure, Cloudflare, Google etc.), there are renewal dates, secrets etc. etc. So the module runs through the code, connects to the APIs of all these resources and checks settings, configuration, deadlines and builds a graph. Has the full picture. Periodically the module runs through the graph and connects and checks for any changes and also runs through the code to spot anything not in the graph etc. So it's a permanent monitoring operation. The landscape graph both defines everything in play but also serves as the basis for ensuring nothing is overlooked as things changed.

[ToolkitSoft® @ToolkitSoft]

@eclectic_dev @badlogicgames LLM agnostic architecture. Then you don't care.

[ToolkitSoft® @ToolkitSoft]

@badlogicgames Scratch humble. That was a typo.

[ToolkitSoft® @ToolkitSoft]

@badlogicgames Folks may I humbly suggest a big fat deterministic software harness managing state locally and using stateless prompting for the LLM might be of value. It's the internet, I could be both right and wrong.

[ToolkitSoft® @ToolkitSoft]

@VadimStrizheus I use a software harness so as much is done without consuming tokens. Makes a big difference.

[ToolkitSoft® @ToolkitSoft]

@TNCYCNSZ @VadimStrizheus The cost difference is enormous, so that's not really a solution for most.