Agents made my retro tech safe to use again and showed their real value as testers of ideas

Once a hoard of outdated electronics, my home now teems with life anew, thanks to the aid of agents. Once considered useless relics, these machines have been rejuvenated and put to work, their potential unleashed through the power of agents.

A 15-year-old device, brimming with an ancient CPU and antiquated software, was once a victim of its own obsolescence. Yet, with the help of an agent, it was transformed into a functional unit, capable of connecting to the modern web. The agent, despite the daunting task, managed to compile the latest cURL with modern cryptography, albeit at a glacial pace due to the device's limited floating point capabilities.

The agents' role extended beyond my own devices. A 12-year-old iMac Pro, long past its prime, was resurrected and running the latest Ubuntu with all necessary firmware tweaks. A virtual reality PC, once underpowered and out-of-date, was also brought back to life. Even a Surface Go, shackled by the limitations of Windows 10, was given new life and utility.

The agents' capabilities have not gone unnoticed in the scientific community. Four key individuals from Google DeepMind recently left their positions to launch Discovery Loop, a venture that applies similar practices to a range of fields including biology, physics, and materials science. The process involves scientists proposing an idea, then setting the agents to work, allowing them to explore multiple paths and arrive at a satisfactory result.

However, the true value of agents lies not in their speed and scope, but in their ability to highlight the real bottleneck in any process - the judgement required to decide what is best. This judgement is a human trait that cannot be replicated by an agent. While agents can automate tasks and provide speed and scope, the human element is crucial in determining which measurements matter, which can be exploited, and which should be disregarded.

The future, as Discovery Loop demonstrates, is not just about automation, but about understanding and applying human judgement to the results.