Engineering Ideas

This post continues the series in which I apply John Doyle’s architecture theory and the hourglass model from network systems engineering to AI agency architecture. Below, I’ll use the terms level, (abstraction,model, theory), component(subsystem, layer), diversity hourglass, composability, hijackability, and others with specific technical meanings described in the previous post in this series.

"Stop Prompting, Start Engineering: 15 Principles to Deliver Your AI Agent to Production" by Vladyslav Chekryzhov deserves far more attention than it has received. As a practicing AI engineer, I can tell the article is born from hard-won experience in production. The advice and checklists in that article are very worth following.

I believe that the most important factor in whether our AI future goes broadly well or poorly is whether people quickly develop effective AI-ready (and AI-enabled) institutions and networks. In that, I agree with the recent Séb Krier's essay "Maintaining agency and control in an age of accelerated intelligence".

Everyone is talking about AI agent architectures, frameworks, and protocols at the moment. Let me apply John Doyle’s architecture theory lens to this topic, as well as Micah Beck’s hourglass model (2019). In this first post in this series, I present the key ideas from the architecture theory and the hourglass model. In the next post, I will apply these ideas in the domain AI agents. The two core concepts in John Doyle’s architecture theory are levels and layers.

This post is a reply to Eugene Kirpichov's post on Linkedin.

This article is the ultimate one in the five-piece series: 1. “The future of OLAP table storage is not Iceberg” argues for why object storage-based open table formats: Apache Iceberg, Apache Hudi, and Delta Lake, although they may completely cover all analytical querying needs for some data teams, impose several significant limitations and inefficiencies for some OLAP use cases, and therefore shouldn’t be trumpeted as the “future” of columnar table storage. 2.

In the previous article, I argued that there is an opportunity for creating a new family of protocols: table transfer protocols, namely Table Read and Table Write protocols to address the M x N interoperability problem between OLAP, timeseries, vector, and search databases on the one hand and data processing/query and ML engines on the other hand.

Software engineers often gravitate towards projects with minimal non-software components and minimal direct interaction with the real world. This tendency, akin to searching for lost keys under a streetlight, stems from their desire to avoid real-world bottlenecks and business risks that can leave them feeling idle or powerless to influence the fate of the product they are working on.

This post is primarily written by Rafael Kaufmann, my contributions were minimal. Warning: a long read! In our first LW post on the Gaia Network, we framed it as a solution to the challenges of building safe, transformative AI. However, the true potential of Gaia as a “world-wide web of causal models” goes far beyond that, and in fact, justifying it in terms of its value to other use cases is key to showing its viability for AI safety.