Lately I open a new app, see they have "added AI" and I already know what I'm going to find: a little bubble in the bottom right corner that opens a chat. I type, wait, receive a generic paragraph. I close it. I go back to doing the task the way I had always done it.
It's not that chat is bad technology. It's that it has become the default shortcut for product teams who want to say "we have AI now" without asking the hardest question: what real user problem do we solve better with this? I've been seeing the same pattern repeat across different products for months, and I think it's worth calling out.
1. Starting with the interface, not the problem
Chat is easy to sell internally. A stakeholder understands it in five seconds, you can do a quick demo and it "looks like real AI". But that means the starting point was the interface, not the job the user needs to get done.
When the order is correct, the question is not "where do we put a chat?" but "at which step in this flow does the user waste the most time or make the most mistakes, and how do we solve it with less friction, not more?". Sometimes the answer is a chat. Most of the time it's autocomplete, a proactive suggestion, a silent automation that you don't even notice is there.
2. Confusing "conversational" with "useful"
There's an established idea that if something converses with you, it's intelligent. But a well-designed form with three fields can solve in ten seconds what a chat takes three minutes of back-and-forth explaining what you already knew you wanted.
At Venturest we have discussed this quite a bit when designing Meetings: the temptation to put in a chatbot that "helps you" is always there. What really moves the needle is taking the transcript, extracting the action points and leaving them already drafted where the user is going to look for them. Without asking anything. Without them having to start a conversation to get it.
3. Not designing for error
A chat hides hallucinations very well. The text comes out fluent, confident, well-punctuated, and the user has no visual cue of when they should distrust it. Instead, when the AI is integrated as an editable suggestion, as a draft or as one option among several, the user maintains control and the error is corrected in the same gesture of accepting or discarding.
Many teams launch the chat without asking themselves what happens the day the answer is incorrect. And in a B2B product, that day arrives quickly.
4. Treating AI as a feature instead of a layer
Another frequent mistake: AI is treated as just another module on the roadmap, something that is "added" to an already finished product. But the AI that truly changes a product is usually not visible as an independent feature. It is woven into the product: in how a list is prioritized, in which fields are autocompleted, in which alerts are generated on their own.
When the user has to "go use the AI" instead of simply noticing that the product now does something it didn't do before, we have already lost part of the value.
5. Not measuring what really matters
Many teams celebrate chat usage metrics (messages sent, sessions, conversation time) when those metrics can mean exactly the opposite of what they think: if a user needs to send five messages to get what they wanted, that's not engagement, it's friction disguised as success.
The metric that matters is whether the task was completed faster, with fewer steps or with fewer errors than before having AI. If that metric doesn't improve, it doesn't matter how many messages have been exchanged.
So, what do we do instead of a chat?
It's not that chat is forbidden. There are cases where it makes sense: when the task is genuinely open, exploratory, without a previous structure. But for most flows in a digital product, there are formats with much more potential and much less design laziness:
- Inline suggestions, where AI proposes and the user decides with one click
- Silent automations, which act without asking permission for low-risk tasks
- Summaries and extractions that appear right where the user needs them
- Intelligent prioritization of what the user sees first, without them having to ask anything
None of these formats are as easy to show in a five-minute demo. But they are the ones that truly change how it feels to use the product.
The most powerful AI I've seen in products isn't noticeable. You notice its absence when you take it away. And that is a much more honest measuring stick than how many conversations were opened this week.
What has been your experience? Have you seen any product that has integrated AI in a way that truly surprised you, beyond the typical chat?
