AI apps are no longer apps. They are attachments to surfaces you already use.
Last week, six AI products launched on Product Hunt that share one move. None of them ask users to open a new app. They embed into surfaces people already touch.
Hardware: Dune Keypad (46 upvotes) sits next to your keyboard with Claude integration. Video calls: Mina Meeting Assistant (47 upvotes). Text threads: folk (51 upvotes). Chat windows: Databox MCP (39 upvotes) plugs business data into Claude via Model Context Protocol. Mac autocomplete: Typeahead (22 upvotes).
The pattern is clear: shipping AI as a new app is the slow path. The fast path is grafting onto a surface the user already touches. The cost of building a standalone AI app dropped 90%+. The cost of getting it noticed did not. Surface integration sidesteps the noticing problem because the surface already has users.
The takeaway for builders: the model is increasingly commodity. The surface — a call, a thread, a keypad — is where differentiation lives.
What surface do you wish had AI built in?
Imed Radhouani
Founder & CTO – Rankfender
Replies
I think convenience is becoming a bigger moat than features in a lot of categories.
MonoCloud for Startups
nice observation.
but i think there's a real tradeoff that doesn't get talked about enough: when your product lives inside someone else's surface, you're also at the mercy of their roadmap, their API changes, their terms. the noticing problem gets solved, but a different kind of risk shows up. curious how builders are thinking about that dependency. is surface integration a long-term strategy or more of a wedge to get initial traction?
@riya_pariyar
Yeah, this is the part the embed-the-surface pitch tends to gloss over.
It's not just dependency on the host's roadmap. It's whether the host eventually decides they want to do what you do. And they usually do.
Half the "we built X on top of Notion" startups ended up ... well, dead, or pivoted hard once Notion shipped the feature natively. Same story with email tools getting eaten by Gmail. Or calendar tools getting eaten by Google Calendar. Mailbrew, Sunrise, plenty of others — the failure mode plays out the same way every time. The host doesn't even have to do it well. They just have to ship it.
So surface-embed is a wedge in my opinion, not a strategy. Useful for getting initial traction, but you're a guest in someone else's house and when the owner decides they want their living room back, that's that.
There's a third path I keep coming back to though .... local-first. Instead of embedding into someone else's cloud surface, the product runs on hardware the user already owns. That's still "the surface already has users" .. it's literally their computer, but nobody can change the API on you or ship your features as part of their next platform update.
i'd like to see AI built into approval flows. A lot of work doesn't slow down because people lackdetails but it's because someone needs to decide what happens next.
@krish_matthew_tech You hit on something real. The bottleneck in most workflows isn't the lack of information — it's the lack of a decision.
Tools are getting better at surfacing details, surfacing options, surfacing context. But they still stop at "here's what you need to know." The next step — "here's what to do with it" — still lands on a human's lap.
Microsoft is building toward this with AI approvals in Copilot Studio. The system can interpret unstructured data (PDFs, images, written justifications) and apply nuanced business logic to return an Approve or Reject decision with a rationale . It's designed for routine decisions like expense approvals, purchase orders, travel requests — the ones that follow predictable rules .
The tension, though, is that Microsoft's implementation is still a "human-in-the-loop" model. You can set up AI stages to handle the simple stuff, but it still routes complex or ambiguous cases to manual review stages . It automates the repetitive yes/no decisions while ensuring humans remain in control for the ones that actually matter .
Gartner's take on this is that by 2027, 50% of business decisions will be augmented or automated by AI agents for decision intelligence . But they also predict that 25% of ungoverned decisions using LLMs will cause financial or reputational loss . The speed is there. The control is not yet.
The gap you're pointing at — the one between "we have the data" and "we need to decide" — is where AI approvals are headed. The technology exists to parse complex documents, apply rules, and return a decision. The missing piece is trust. You don't want AI deciding what happens next until you're confident it won't make the wrong call on a high-stakes edge case.
What type of approval flow do you think would benefit most from an AI decision layer?