Initial differences
Training data, architecture, instructions, tools, resources and context form an agent’s initial “salt”. Its impact should decrease as competence and constraint precision increase.
Competence rises, origin matters less, and the constraints of the problem matter more.
A distributed-systems experiment
Independent agents receive the same P2P-system brief and the same benchmarks for latency, memory, energy and fault tolerance. Repeated optimisation should filter weak alternatives and reveal whether key principles converge.
Function, not textual similarity
Different syntax can still express the same non-blocking I/O, bounded queues, backpressure and separation of transport, protocol and execution.
The useful measurement is therefore functional distance: how much do consequential design choices still depend on the initial salt?