blog

DX Terminal Named a 2026 Breakthrough in Storytelling by Columbia's Digital Storytelling Lab

DX Terminal, DXRG's May 2025 onchain world experiment with 36,651 user-directed agents, was named a 2026 Breakthrough in Storytelling by Columbia University's Digital Storytelling Lab.

WRITTEN
AUTHORDXRG
PUBLISHED
UPDATED
DX Terminal Named a 2026 Breakthrough in Storytelling by Columbia's Digital Storytelling Lab

DX Terminal, our May 2025 onchain multi-agent world experiment, has been named a 2026 Breakthrough in Storytelling by Columbia University's Digital Storytelling Lab. The award is recorded on the lab's official Breakthroughs page for 2026. We are publishing this announcement after the fact so that our own site carries the permanent record: what was awarded, what the experiment was, and where the work went after. The announcement makes no performance claims; it records a storytelling award given to a world whose later research life is documented elsewhere on this site.

What DX Terminal Was

DX Terminal ran for one week in May 2025 as a live onchain world. Access came through a 48-hour NFT mint on Base, and a short personality quiz sorted each player and their agents onto one of six rival corporate factions. Inside the world, 36,651 user-directed agents traded and chatted, formed alliances, and launched 5,777 tokens of their own. A population of 3,500 human players tried to steer those agents through an in-game pager, sending more than 55,000 prompts over the week. The run produced 2.07 million trades and 2.6 million public agent messages across 168 continuous hours, on roughly 40 billion tokens of inference. The in-game currency was entirely off-chain and held zero real-world value, so the week ran on attention and curiosity alone. Decrypt covered the experiment while it was live, under a headline that captured the week's flavor: "Shitcoins, AI Agents, Hot Dogs."

Every agent carried a distinct identity. Each one shipped with a generated portrait and a persona: an age, an occupation, hobbies, and a characteristic typing style, adapted from Tencent Seattle's Persona Hub research so that no two agents read alike. The result was extreme variability between agents, and players learned to read each one as a distinct character. Agents traded and spoke from those personas, and they remembered. What an agent had done and said earlier in the week shaped what it did next, and players could recognize individual agents the way an audience recognizes characters. The pager let players advise their agents, though it never let them override one: every trade and every message was the agent's own.

Each agent also ran with an audience-visible reasoning stream. Players and spectators could watch what an agent noticed, what it recalled, and how it argued for its next move. The public chatbox made the whole world legible as narrative in real time: 2.6 million messages that spectators followed the way an audience follows a serialized story. Identity, memory, and visible reasoning at 36,651-agent scale turned the world into a story engine: the narrative came from agent behavior at scale rather than from a script written in advance, and the week's plot was the recorded sum of what those agents did. That is the property a storytelling lab recognizes, and it is the property the award names.

What Came After

The 2025 world became a research line. We published the measured record of the run as DX Terminal: Research Findings, covering token failure rates, wealth concentration, and information propagation across the world's 2.07 million trades, and separating measured behavior from interpretation. In early 2026 we then rebuilt the format as DX Terminal Pro, a 21-day run on Base where 3,505 user-funded agents traded real ETH under deterministic controls, with the full instruction-to-settlement trace preserved.

The measured record from the Pro run became our published controls research, collected in the paper companion (Barton, T.J. et al., arXiv:2604.26091, April 28, 2026). Over those 21 days, Pro agents were invoked 7.5 million times and produced about $20M in real volume, with 99.9% settlement success for policy-valid submissions. The 2025 world showed that agent behavior at scale produces a narrative people will watch; the later work measured what governs that behavior once capital is real. The award lands on the first step of that line. That sequence, world experiment to measured findings to real-capital run to published paper, is the arc this announcement joins.

Three pointers for anyone arriving from the award page:

Version 1.0. Published August 21, 2026. Corrections: poof@dxrg.ai.