← Articles 2026-06-08 Deniz

Architecture is All You Need

In Late Autumn of 2025 I was discussing a problem I had with software and agents, particularly in production systems, with some of the quantitative researchers & analysts in my university.

I described a sandbox written in a language such as Rust or OCaml (fuck that) that can assist with token prediction, create coding barriers, encapsulate memory and make merges with production in a live environment.

The question then morphed into something more abstract and philosophical: How do we create ambient software awareness?

A few months go by and I fell into a rabbit hole of deep reading and research into what Developer Tools are missing, and how I can improve LLMs in production code environments.

Dated April 11, 2026, I arrived home from work.

At precisely 17:07:55 I created a new Markdown called Icarus. I put all of my overthinking from my day at work into paper, validated it in a few test environments, taught myself Rust, and discussed with Claude as my design partner.

I identified a gap: No tooling existed that maintained a live, queryable and agentic model of your codebase.

My stripped-down thesis is as follows:

Make the thesis into an engine of behaviours and defaults, and you have NEAT.

I aimed to start development in September of 2026, after I learned Rust to where I could create the compiler and home-brew tree-sitter to a standard of my liking.

In the last 19 days, I've seen great advancements in architecture-based programming tooling, after the advent of Andrej Karpathy's Gist. My biggest regret is not having seen this tweet when it came out.

However, it validated and democratised the concept of architectures in the wider open source community. I am eternally grateful for it, and to me, it shows that my level of thinking as a CS Undergraduate is along the right path.

I renamed the project to NEAT (Networked Environment Architecture Tools) as the scope solidified into its methodologies. By no means is the architecture the feature.

On day 18 I was studying and found several advancements within the industry. My close friend, now co-founder of NEAT, warned:

"You can do it once and do it right, but if you take your time you might not be able to do it at all."

So development of the MVP in TypeScript starts on April 30, 2026, documented above. Time: 15:45:25 GMT. The perfectionist in me disagrees, but this is what co-founders are for. One must be objective in such stakes.

I'm in this alone until v0. At v0.1, I will be accompanied by my co-founders. They where set in stone long before this project initiated.

We began university making websites for small businesses, and grew together. They're the best devs i know.

By no means is the graph "new art." In the current open source landscape, it has become commodified. We have tools from both open source startups and big tech who either adapt the architecture model, symbolise it, or use it one way or another within their toolkit to improve LLM accuracy.

For NEAT, the architecture is simply the petri dish. The landscape. The fabric. In a NEAT world, repositories are live by default, and the truth is non-negotiable.

Follow my GitHub to see what I end up pulling off. https://github.com/dcdeniz