A software harness using stateless prompts trumps an LLM harness (Faster, cheaper, better outputs).

No skills. No markdown. Faster, cheaper and better outputs. LLM agnostic and cross platform. I use a deterministic software harness. You should too.

Finally some stuff I've been working on for a long time starts to come together. It's made up of 4 different things that I vibe coded in 20 minutes on my phone. There is a methodology which defines the principles of how to work effectively and safely with AI called Systemorama® There is an AI focused language (asp like syntax that transpiles to C#) called Progorama® There is a framework which has the Promptorama® engine. There is a desktop application called Promptorama® There is usage of knowledge graph principles called Synthorama®. It's an architecture that is LLM agnostic, cross platform (.net cross platform and electron for the UI). I even wrote a language with a defined parser and lexer. No skills are used (they are terrible). It's highly deterministic. It's based on System Engineering principles. It's a deterministic system controlling a non-deterministic component (the LLM). It's very token efficient. I'm still making improvements and improvements, but already I start to see it's efficient. A very important part of it is that all prompts are stateless by design. They standalone, an island of logic so to speak. All inputs and LLM outputs are structured. All usage of an LLM is essentially a function call by the deterministic workflow. I have not written any code directly this year. It's nuts. However I spend hours studying what is happening and have workflows for defining the work to be done. I take a very spec driven approach. By design anything that can be done without an LLM is done by the software harness. That turns out to be an awful lot. This of course means my LLM usage of quota is rather low. I tend to use Claude and Codex together. My preference is for implementation by Claude if possible (depending on quota). Claude is always the supervisor. Up to a few months ago I used Gemini a lot. I will be using Google Argon when it comes out. I expect it to be very good. What really ads value is I always run the Reviewer on a different LLM to the implementer. The sidekick is a second reviewer and I run it on the LLM that carried out the implementation. So there is a a lot of review and it is very valuable. It's very cool to observe the 4 sessions "talking" to each other. The golden rule of the Systemorama® methodology is that the LLM can only propose a change (to anything not just coding), the software harness validates the LLM proposals in every way imaginable (even satisfiability modulo theories are used). That principle is applied all the time. When the LLM proposes the change overlays are used to "simulate" the change if it was made and a full compile is made and validates. This flushes out issues that must be addressed. 0 compile warnings or errors is enforced. All bug fixes require a failing unit test to be written if unit coverage was missing and it must fail before the fix is made. So there is a constant improvement in coverage to prevent regression. I will be commercialising this but I'm also a developer at heart, so have the intent to make it highly accessible. There are many different ways to commercialise. I'm trying to avoid the VC route. Still figuring it out but have some good ideas. I would like to acknowledge Innoviris in Bruxelles who gave me a subsidy and were very helpful. ToolkitSoft is right in the heart of the EU in Bruxelles. It received a subsidy during this development from Innoviris. Here's the thing about Europe. Companies like mine get a lot of help from both the Belgian government and the EU. A few very quick points. It coordinates multiple LLMs working together on the same work load. It is a software harness deterministically controlling the whole work being implemented. State is kept locally. Zero vendor lockin. Stateless prompting is used. It is considerably more ecological than the typical swarm of agents approach being peddled. It's inherently safer.

Hola @CaolanReports Due to my incessant embarrassment over those Enterprise Ireland approved supply chains that feed the Russian production of those bombs that kill innocent Ukranians. I have an idea. Why don't you start a gofundme to raise money to put simple full page ads in the national media or on TV in Ireland. To state simple facts about what is happening. Ireland is curently complicit in the murder of innocent civilians in Ukraine. I think you should do this. #guaranteedirishbombs

AI gives us leverage, but the intelligence still comes from humans.

independent.co.uk/news/uk/ho… Mike's cholesterol will be ok after the apocalypse.

I'm a small business, the type I'm always hearing you guys love. I want to integrate invoices into my backend software automatically. So I can be fully automated. What it has done is made my life hard administratively speaking.