Sep 8, 2025
Why automated discovery matters for energy systems
A short argument for using automated search to find deployment paths and system designs that are easy to miss manually.
Searching The Design Space
Energy systems contain too many interacting constraints for manual search to cover well. Automated discovery helps map the space, but its value depends on asking questions that preserve real-world constraints.
The point is not to produce a single answer. It is to find clusters of promising directions, identify sensitive assumptions, and reveal options that would otherwise stay hidden. A discovery tool that cannot explain the shape of the search space is not useful enough.
Capturing Useful Surprises
The best discoveries are often near the edge of an existing intuition. They look strange at first, then become obvious once the mechanism is visible. Automated systems can help find those edges because they do not get bored searching unglamorous combinations.
But surprise is not a metric by itself. A surprising candidate has to be connected to a mechanism, a test, and a deployment constraint. Otherwise it is only novelty.
Where Human Judgment Belongs
Researchers should not be removed from the loop. They should be moved to the parts of the loop where judgment matters most: defining constraints, inspecting failure modes, deciding which uncertainty is worth spending lab time on, and rejecting candidates for reasons a model cannot know.
The best automated discovery systems make those judgments easier. They provide structured evidence, not authority. They help a lab ask better next questions.
Energy Lab Criteria
For Monolith, an automated discovery result must connect to one of three outcomes: a simulation to run, an experiment to perform, or a deployment decision to clarify. If it does not produce one of those outcomes, it stays in the research backlog.
That criterion keeps the lab grounded. Energy research is full of interesting abstractions. The useful ones eventually touch a material, a device, a sensor, a controller, or a field decision.