AutoResearch searches program space, not a configuration grid. Here's the difference.
With AutoML, you define a configuration space and the system searches it. You specify ranges for learning rate, batch size, and layer count. Neural Architecture Search (NAS) does the same for architecture patterns. Both are limited to options you define upfront.
With AutoResearch, your agent edits actual Python code. It can swap the optimizer algorithm, rewrite the attention mechanism, or restructure the training loop. Your search space isn't a grid of preset values. It's arbitrary code.