A new preprint on arXiv, titled "The Wealth of Agents: Scaling Agentic AI Across Developers," identifies a bottleneck in current agentic AI development. According to the abstract, today's agentic systems are usually created and operated by a single developer who controls every participating agent. While that makes development straightforward, the authors argue it restricts the scale such systems can reach.
The abstract does not specify what the authors propose instead, as the text cuts off after the word "We…". Still, the framing suggests a shift toward distributing agentic AI across multiple developers or teams. Because this is the only source, there is no independent comparison to make; the article simply reports the paper's stated motivation.
If the full paper delivers on its title, it could point toward a more collaborative model for building large-scale agentic systems. For now, the key takeaway is the identified limitation: full control by a single developer may be the very thing holding agentic AI back from scaling further.