Modern agentic engineering seems to be converging on a theme of agents being exceptionally great at taking care of boilerplate and navigating large search spaces, pursuing an objective function over long time horizons. They are much less great at decision-making and architecture design. The net effect is a speed-up of the software engineering projects with humans still being essential parts of the loop. The human programmer’s ability to effectively work with information is becoming a bottleneck to realizing the system’s potential.

In this article, I’m sharing a few mental models one can use to optimize their information processing capability as a human in a human-agent engineering system, to an effect of spending limited human attention capital more efficiently.

The key to an effective AI software engineering flow is bureaucracy. Everything needs to be indexed, named, and organized.

As I wrote in the previous article, the developments with Cursor gave rise to an opportunity to replace spreadsheets with React apps, while saving on time and brainpower. In this article, I’ll share more tips on what happens after the app is developed. Inevitably, you will discover bugs and room for improvement for it to match your vision precisely.

Recent developments in Cursor AI-powered IDE have effectively made it so that it takes less time and effort for me to develop a React app for my administrative needs rather than turn to a spreadsheet. This post describes the technique I use to do so.

Programming is easy and effortless when things go right. This happy path is where AI shines. There are sandboxed environments of development, coheasive ecosystems of tools, frameworks, libraries, which play well together to achieve a certain goal. There are also generally accepted patterns and ways of doing things. While you’re working within such a sandbox, exactly following the purposes it was built for, you’re on the happy path.