As reported by MIT Tech Review AI, the core question here isn't whether the agents behaved in interesting ways, but what that behavior signifies for learning. The finding that 'playful' agents later performed better on specific tasks is the most concrete takeaway, yet it remains unclear how scalable or generalizable this approach is beyond a tightly controlled sandbox. In our view, the experiment serves less as proof of AI play and more as a useful prompt for the field to consider how unstructured exploration could complement traditional training paradigms, especially for systems destined for unpredictable real-world environments.