Principles
As more work is delegated to AI, it becomes increasingly important to preserve human decision-making, check results, and recover when problems occur. Root Kernel’s five principles guide how we put these needs into the design, validation, and operation of our products and tools.
Define the specification before starting work.
Set out the goal, scope, and completion criteria in a specification. Use it as the basis for writing code and carrying out the work.
Build an environment where results can be reproduced and checked.
Keep the conditions and execution records needed for validation so the process behind a result can be checked again.
Give AI room to act autonomously, with people making consequential decisions.
People define the intent and boundaries. AI agents work within them and review one another’s work. People give final approval.
Validate before advancing, and keep a path back.
Check the conditions for advancing at each stage. Record task state and the reasoning behind decisions so it is clear where to return and what to repeat if a problem occurs.
Improve technology through problems found in products.
Use problems found during product development to improve the technology, then apply those improvements back to the products.