“Can you check the customer enquiries and prepare a follow-up list?”
It sounds like the perfect task for an AI assistant. A little reading, some organising, and a useful report waiting for your sales team.
Then comes the awkward part:
The enquiries are inside your CRM. The assistant needs to sign in. Would you give it your password?
On 22 October 2026, 1Password announced a platform designed to address that problem. With supported integrations, AI agents can request a login, receive your approval, and access an account without seeing the password itself.
That could make assistants more useful. It also gives business owners a new decision to make.
Think about onboarding a new employee.
You might give them access to customer enquiries. You probably would not immediately give them permission to export every customer record, change prices, and send quotations without review.
An AI assistant needs similarly clear boundaries.
Imagine a Bengaluru service company receiving customer enquiries throughout the day.
An assistant could prepare a morning summary showing:
- New enquiries received since the previous close of business.
- Messages waiting for a response beyond target SLAs.
- Missing details the sales team needs to request (e.g. project scope, budget).
- Draft follow-ups ready for a human team member to review and approve.
That is a focused job with a concrete result someone can check.
Giving the same assistant unrestricted permission to contact customers directly would be a much larger decision.
Before handing credentials or API tokens to any autonomous tool, establish these boundaries:
| Boundary Question | Operational Standard |
|---|---|
| 1. What does it need to read? | Give it access strictly to the information required for the immediate task—nothing more. |
| 2. What is it allowed to change? | Reading records and editing them must be separate permissions. Default to read-only. |
| 3. Which actions need approval? | Sending outbound messages, issuing quotations, and processing payments require explicit human sign-off. |
| 4. How do we stop it? | Your team must have a one-click revocation protocol and clear activity audit logs. |
These questions remain essential even if the tool promises state-of-the-art encryption.
Choose a task that is repetitive, easy to check, and unlikely to cause harm if the first draft is imperfect.
Preparing an internal summary is a reasonable starting point. Making legal or financial commitments to a customer requires substantial guardrails.
Run a small pilot. Review the output. Measure whether it genuinely saves staff time before expanding access.
The exciting part of AI is what it may help your team accomplish. The useful part is making that help dependable.
