BlackArcAbout
A field guide to how agents get built
The Agent Atlas is a close reading of the research behind modern AI agents, turned into one browsable publication for anyone building with AI: shipping applications, automating processes, running a business, building harnesses, or refining how they work. It is built and maintained by BlackArc Industrial, which builds production AI agents for manufacturing and operations.
- Papers
- 162
- Topics
- 15
- Reading
- 47h
Why The big idea
Most people read the abstract. Builders need the mechanism.
Every entry goes past the summary to the part that matters when you sit down to build: what is genuinely new, how it works down to the math, pseudocode of the core idea, and where it breaks. Then it maps the idea onto your own work: the applications you ship, the processes you automate, the business you run, the harness you build on, and the way you and your agents work day to day.
Method
How it was made
Read the paper in full
Each source is read end to end, not skimmed. 133 of the 162 entries link straight back to their paper on arXiv.
Write for a builder
Plain language, the math demystified, a "How You'd Use It" section that maps the paper onto your own applications, automations, business, harness, and workflows, and a path to a version you could build yourself.
Map it into the atlas
Every breakdown lands in one of 15 topics, so the collection reads as a single publication instead of a pile of PDFs.
Go further
Two doors from here.
BlackArc builds the agent systems these papers describe. Josh writes about the ideas and the building. Both are one click away.