Consider a team whose contact records for a key client exist in three places: a phone's address book, a shared spreadsheet, and a chat thread where actual conversations happen. All three hold different values. One has an old phone number. One has a misspelled email address. One is current. No one on the team knows with certainty which version is correct.
The problem sounds like a minor inconvenience. The actual cost is visible in what happens around it. Every time the team's AI assistant needs to reach that contact, it has to choose which source to trust, and there is no principled basis for that choice. Every project involving the contact starts with informal reconciliation: which email is live, which number is active. Any system built on top inherits the confusion, the way a building inherits a crooked foundation. And the problem does not stay contained to contacts. Deal values recorded in one tool, deal status updated in another. Tasks tracked in two applications that show different states. Dates in messages, in calendars, in people's heads.
The instinct is to solve this with effort: update everything everywhere, be more careful, enforce a convention. That is treating a structural problem as a behavioral one, and the behavioral fix does not hold. The structural fix is simpler and more durable: every fact gets exactly one home. One location where the authoritative version lives. Every other system reads from that home and writes back to it. Copies stop being rival versions and become what they should always have been: views onto a single record.
When teams apply this principle to a genuinely messy area, an entire category of work tends to disappear. Not decrease. Disappear. The reconciliation step at the start of each project goes away, because nothing has drifted from anything else. The disputes about whose version is correct go away, because there is only one version. The assistant stops having to choose between sources, because there is only one source to point it at.
This is what the phrase single source of truth means in practice. It is not a rule about tidiness. It is a structural decision: choose where each fact lives, and treat everything else as a derived view. A view can be read. It should not be written to directly. When a copy is written to as though it were an original, it becomes a rival original, and the problem returns.
This lesson belongs in a program about AI for a specific reason. An assistant is only as reliable as the ground it stands on. Point one at three contradictory records and you get a confident amplifier of contradictions: outputs that are precise in form and unreliable in substance. Give it one source of truth and its precision becomes trustworthy. Structure is not housekeeping. It is what makes delegation safe at all.