Answers
One rule, four prompts: why your AI automations drift apart
When the same rule is pasted into four automation prompts, you do not have one rule — you have four copies that will stop agreeing. Correcting it means finding every file that carries it, and the copy you miss keeps running. With Arroway the rule is written and approved once, in a shared place outside the prompts, and every routine reads it when it starts. The correction is a single edit, and no prompt is left holding a version nobody remembered to update.
Last updated September 3, 2026
The rule is not in one place, so it is not one rule
It starts reasonably. A routine needs to know the cut-off, the exception, the step that requires a person, so you write it into the prompt. Then a second routine needs the same thing, and copying it is faster than designing anything. By the fourth, nobody knows which copies exist. When the rule changes, three are updated and one is not — and the one that is not does not fail. It runs, confidently, on a rule the team retired.
What the prompt keeps, and what it stops keeping
The prompt keeps the procedure: what this routine does, in what order, where it delivers. The rule moves out — the cut-off, the exception, what needs a person, what the team decided and the reasoning it cannot be applied without. It becomes one entry in Arroway, approved by a person, read by every routine that needs it. Nothing has to be remembered at the right moment: a routine reads what applies when it opens, and nobody has to go anywhere to look it up.
A dead fact survives inside the copy nobody maintains
A duplicated rule does not only diverge, it fossilises. A schedule that no longer exists, an exception that was revoked, a step that was removed: each keeps being quoted by whichever copy stopped being maintained, and being quoted looks exactly like being right. In Arroway each entry is stored with the condition that ends it, so a superseded rule stops being served and the replacement is what routines read. The old wording stays on the record, in one place, where it cannot be mistaken for current.
What it looks like in practice
A team runs four scheduled AI routines. Three were given the same rule about which changes need a person to sign off; the fourth was written from an older version of it. Two of the prompts still quote a time grid that was replaced in June. Nothing breaks. The routine with the older rule keeps skipping a sign-off nobody meant to skip, and the two with the dead grid keep citing hours that mean nothing — and all of them sound exactly as certain as the ones that are right. Written once instead, the rule is a single entry with an owner and a review date: the correction is one edit, and all four routines read the corrected version on their next run.
Questions people ask about this
- Isn't this just a shared config file the prompts import?
- A shared file solves the copying and leaves the rest. It has no step where a person approves a change, so whatever is written into it governs immediately. It has no expiry, so a rule that stopped being true stays in force silently. And it does not say who decided what, which is the first thing anyone asks when two rules conflict. Those three are the difference, and they are what a rule needs and a file does not have.
- What if a routine ignores what it reads?
- It can, and it is worth being exact about this. A routine reads the rules that apply when it starts — nobody has to remember to fetch them and nobody has to open another tool — and how firmly that reading can be required depends on the AI tool the routine runs in. What covers the gap is that the reading is visible: you can see what was served and what was written back, and a short human pass catches a run that acted against a rule. It is not a lock, and a page telling you it was one would be lying to you.
- Who keeps the rules from becoming their own mess?
- Someone does, and it is worth saying how much. The writing happens as a side effect of the work: the routines record what a run settled, as they finish. The upkeep is a short pass over a list Arroway puts together — what an AI proposed and nobody confirmed, what has reached its review date, what no routine has read in a month, and any two entries that contradict each other. Arroway finds those and shows them; deciding what stays and what goes belongs to a person, and nothing is merged or deleted on its own. That is minutes a week, and skipping a week breaks nothing.
Where this is verifiable
Product documentation on this site (How it works, Install), the sanctioned product spec in the repository, and the Privacy Policy for what is stored and who can reach it. Everything described here is behaviour the tools apply today, not roadmap. How firmly a routine's opening read can be required depends on the AI tool it runs in, and this page says so where it matters.
https://www.arroway.app/en/answers/one-rule-four-prompts-and-they-drift