Agentic Tend develops a general agentic context model without encoding intelligence that general models can learn. Software development is its first complete domain realization.

Tend means caring for a growing system: preserve non-inferable objectives and preferences, add structure when evidence calls for it, and let projects retain their own shape.

model capability -> L0 epistemology -> L1 domain realization -> L2 repository facts and contracts -> mechanical evidence

Semantic ownership ends at L2; mechanical evidence is an executable projection of an observable contract.

• AGENTS: unconditional cross-domain epistemology at user scope and repository-specific ambient facts in each project.
• Skills: reusable workflows and domain realizations, including project-context bootstrapping, software engineering, and persistent prose.
• Human docs and decisions: motivation, public theory, and evidence-backed durable rationale.
• Tests, hooks, and CI: repository-owned mechanical evidence added when a reliable oracle exists.
• Generators: explicit materialization justified by repeated deterministic pressure.

• Agentic tooling defines the motivation, semantic model, and activation mechanisms.
• Context ownership routes persistent context to its narrowest sufficient owner.
• Development routing routes contract, planning, durable-work, and execution uncertainty.
• The organization roadmap tracks evidence and extension work spanning multiple layers.