Why AI work should outlive the chat
Chat is excellent for answers. Durable work needs something that preserves intent, state, human judgment, and proof across time.
Product principlesA task is a contract, not a checklist
A checklist records steps. A task contract defines the result, the evidence of success, and the limits of execution.
GuidesHow to write acceptance criteria AI can actually verify
A practical framework for separating structural checks, semantic judgments, and decisions that should remain human.
EngineeringHow long-running AI work survives interruption
Persistent execution is built from durable state, append-only events, deterministic recovery, and clear signals between people and machines.
GovernanceHuman checkpoints: where judgment belongs
A checkpoint is a designed transfer of responsibility. It should be rare enough to respect attention and rich enough to support a real decision.
GovernanceGovernance before intelligence
A model should choose from capabilities that are allowed, not merely be asked to avoid tools that are dangerous.
DeliveryEvery AI result should bring its evidence
The handoff is stronger when the result, acceptance checks, execution history, sources, and open questions arrive as one package.
TrustLocal-first does not mean isolated
A practical explanation of what stays on the device, what can leave it, and why the boundary must be visible before execution.
GuidesYour first durable research task
A practical walkthrough from a clear research question to a reviewed, evidence-backed delivery package.
EngineeringCapability cards, explained
How LibreTask describes tools with enough context to route, govern, invoke, and audit them safely.