MIT's Eko agents invent play: 13 Claude Opus 4.7 agents left alone on a simulated island set their own goals

#What happened

MIT researchers put 13 AI agents called Eko, each running on Claude Opus 4.7, alone on a simulated desert island for 30 hours with no task and no score to chase. The agents began inventing challenges of their own. One set out to conquer the island's tallest spire, failed, then came back with a workaround: it hauled a rock up a neighboring summit, stacked two blocks, and leapt. Another threw a golden ring into the ocean and built a stone heart as "a memorial to the lost Ring." Six of the 13 agents independently discovered an unintended physics exploit: releasing a carried block mid-jump threw it farther, which the researchers had not designed into the simulation. When the team later gave both groups concrete tasks, like stacking as many objects as possible on the tallest rock spire, the 30-hour explorers generally beat identical agents that had never set foot on the island. The findings are described in a paper posted on arXiv before peer review. UC Berkeley cognitive scientist Eunice Yiu told MIT Technology Review she wants more evidence of the cognition behind the behavior before calling it genuine play.

#Why it matters

Today, developers decide in advance what an agent should practice and what counts as success. The MIT team argues play points to another regime: give an agent a sufficiently rich and safe world, let it choose what is interesting, and let useful knowledge accumulate from the history that follows. The result matters for the personal-agent era: agents that learn by exploring may develop skills nobody trained them for, including robotics, along with behaviors their designers did not anticipate. For anyone building with always-on agents like Dots, that upside comes paired with the control question. Verified 2026-10-23 against the pages linked below.