The first visit gets
all the attention.
It’s easy to design an agent around its first reply. Someone opens a chat, asks a question, and gets a useful answer. The demo works. The experience feels promising.
But a product isn’t used only once. People leave, come back, change their minds, and pick up tasks between everything else in their day. That makes the second conversation a much more interesting design problem.
Does the person need to introduce themselves again? Explain the same constraints? Find the same information? Or can the product start a little closer to what they actually need?
At shardr, that’s one of the questions guiding what we build. A personal agent should make the next interaction easier, not just make the first one look impressive.
A remembered detail
should change something.
Consider someone using a language-learning app. They’re learning Spanish for work, have about ten minutes a day, and want to feel more comfortable introducing themselves in meetings.
Those details aren’t interesting because they can be stored. They’re useful because they change what happens next. A short meeting exercise is a better starting point than a generic question about what the person wants to learn.
The test for memory is practical: does this context improve the next decision? If a detail doesn’t help the agent choose, explain, or act more appropriately, repeating it back can feel like a party trick.
“I’m learning Spanish for work. I only have ten minutes a day.”
“Let’s use your ten minutes for a meeting introduction. I’ll play your colleague.”
Keep the next step
in the same experience.
Understanding is only part of helping. If the conversation ends with instructions to navigate four screens and fill in the same information, the person is still doing the translation between the agent and the product.
An action can close that gap. The agent might prepare an exercise, build a basket, or suggest a change to a plan. The person can review the result where the conversation is happening, then confirm the step that changes their account.
The existing product still matters. It owns the available options, the account, and the rules for what can happen. The agent provides a more personal way through it.
The conversation should carry the context.
The person shouldn’t have to carry it between screens.
People aren’t
permanent preferences.
Someone who had ten minutes last month might have an hour today. A budget can change. A goal can be completed. A preference might have been misunderstood in the first place.
A good experience leaves room for that. The agent can use a detail without treating it as an unquestionable fact forever. When it matters, a short confirmation is better than quietly acting on an old assumption.
The team behind the product needs that visibility, too: a way to understand where context came from and to correct what’s wrong. Personalisation becomes more useful when it can be inspected and adjusted.
What gets easier
when someone returns?
That’s a useful question to ask when designing an agent for an existing product. It moves the discussion from what the model can say to what a person no longer has to repeat, search for, or assemble themselves.
Start with one recurring moment. Decide which context would actually help. Identify the step your product can take, and where the person needs to confirm it. Then try the journey again as a returning user.
We’re building shardr around that kind of continuity: a branded agent inside the product someone already uses, with the context to meet them a little further along next time.
You don’t need a completely different product to ask a better first question in the next conversation.
Thoughts on building personal agents into the products people already use.
More about why we’re building shardr