I had a team that held “robot wars” between different bits of code. We had a policy agent that would assess two different robots tasked with the same thing and the policy agent would “kill” the lower performing code and promote the one that was better. I’ve built something similar into my current platform where agents monitor other agents and then assign development agents the work to improve the work.
This is a great real-world example of the idea behind the post. The key isn't that agents write code, it's that the system can observe outcomes, compare alternatives, and feed improvements back into the development loop. That's when an agentic system starts behaving more like an evolving organism than a static automation.
I had a team that held “robot wars” between different bits of code. We had a policy agent that would assess two different robots tasked with the same thing and the policy agent would “kill” the lower performing code and promote the one that was better. I’ve built something similar into my current platform where agents monitor other agents and then assign development agents the work to improve the work.
This is a great real-world example of the idea behind the post. The key isn't that agents write code, it's that the system can observe outcomes, compare alternatives, and feed improvements back into the development loop. That's when an agentic system starts behaving more like an evolving organism than a static automation.