Start where the account already lives.
Your users already have a history with your product: an account, saved choices, unfinished tasks. Asking them to start again somewhere else puts distance between the conversation and the things it could help with.
An agent inside that existing experience can become another way to use it. Someone might browse the usual screens one day, then ask a question the next. Both paths should lead to the same account and the same result.
Keep the product recognisable.
The chat should sound and look like it belongs. But continuity is more than matching a colour. A saved plan should appear in the usual place. A basket built in a conversation should still be editable in the usual basket.
That gives people a choice: talk when it helps, use the interface when it is quicker. The conversation earns its place by making a particular moment easier.
Begin with one useful moment.
Choose a task where people already need help translating what they want into what your product offers. Make the handoff between conversation and interface clear. Then test both directions: can someone start in chat and finish in the app, or return to chat after making a change?
The ambition is a more personal product. A chatbox is the place that experience begins, not the whole experience itself.
Thoughts on building personal agents into the products people already use.
More about why we’re building shardr