The receipt belongs next to the proactive text.
The user should not dig through logs to understand why the agent interrupted. The receipt should fit in a compact detail view or follow-up message.
How to add rule receipts to a text message AI agent
A text message AI agent should not simply interrupt. It should explain why now, which source proved urgency, which rule fired, and how the user can correct future behavior.
Start with the receipt shape before adding more autonomy.
Rule receipts are small, structured explanations attached to proactive texts. They make user trust easier to earn and bad alerts easier to correct.
Trigger
What event caused the agent to consider texting?
Source
Which message, calendar item, browser state, or task proved urgency?
Rule
Which policy decided interrupt, batch, escalate, or archive?
Correction
How should the user score and rewrite the behavior?
Implement the receipt in four passes.
Keep the first version simple enough to review weekly.
Capture the trigger and source.
When the agent sends a proactive text, store the event and proof. Link the text to its source evidence.
Attach the policy version.
Record whether the agent chose interrupt, batch, escalate, or archive. Include the policy wording that made that choice.
Close the correction loop.
Let the user mark the text useful, early, late, duplicate, unnecessary, or missed. Feed that back into the agent prompt.
Use the same receipt across agent channels.
SMS is the best place to start because attention cost is high. The same pattern can later cover browser work and generated deliverables.
Trigger proof
Show why the agent considered the event urgent.
Policy choice
Explain why text was chosen instead of digest or archive.
Channel decision
Use the same choice framework for computer-use cache workflows.
Correction review
Preserve feedback when an AI agent builds websites.
Implementation checklist
- Add a receipt ID to every proactive text.
- Store trigger, source, policy version, channel choice, and confidence.
- Show the receipt in a compact user-readable format.
- Add one-tap review labels: useful, early, late, duplicate, unnecessary, missed.
- Convert review failures into system prompt and policy updates.
- Review receipts weekly before expanding autonomy.
Prompt clause
Before sending a proactive text, create a rule receipt with the trigger, source, rule, channel choice, and correction path. If you cannot explain why the text is necessary now, batch the item instead.
Sources
- Microsoft Research on interrupted work for attention-cost framing.
- Nielsen Norman Group on notification design for alert clarity and user-control principles.
- Supers for personal AI agent workflows across SMS, browser work, and generated outputs.
FAQ
Should receipts be shown in every SMS?
The text can stay short, but a receipt should be accessible whenever the agent interrupts proactively.
What is the minimum useful receipt?
Trigger, source, rule, channel, and a correction path.
Can receipts reduce alerts?
Yes. Review scores reveal rules that send too many unnecessary texts.
How does Supers fit?
Start with the text message AI assistant, then reuse receipts across other Supers workflows.
Make every proactive AI text explain itself.
Supers can help teams test personal agents that text, browse, and build while keeping rule receipts visible and correctable.