Assistance and authority are different jobs
AI can help organize notes, compare explicit fields, prepare a draft, format a reviewed decision, or check work against a list. These tasks can remove repetition without transferring authority.
Authority stays human when the work approves access, makes a promise, interprets unclear evidence, handles sensitive context, chooses an exception, or affects another person in a consequential way. Fluency does not turn generated output into a responsible decision.
Design the review before using the tool
“A human checks it” is not a complete control. The reviewer needs to know which source to compare, which facts must never be inferred, what counts as an unacceptable change, and when the task must return to a manual path.
A bounded test should include ordinary, incomplete, and unusual examples. Record incorrect additions, missing facts, ambiguous wording, and the time required to verify the result. Speed matters only when the reviewed work remains useful and truthful.
Keep failure visible and reversible
Use the least information and access the task requires. Preserve the source material. Name the person who can stop the assistance. Keep a manual fallback for outages, inconsistent output, or cases the tool cannot handle responsibly.
MethodCo uses “founder-led and AI-assisted” because responsibility is not automated. Claims, terms, quality, privacy, publishing, service recommendations, and product decisions remain owned by a person.
Source notes
- Moral ConstitutionThe public rules for truthful claims, privacy, quality, and accountable AI use.
- Practical AI literacy pathA free workflow-review path for applying the boundary to one real task.