You were never paying for the software.
Last week I sat in a Berlin startup office with thirty Airbnb hosts, wine, pizza and one piece of software that every person in the room had chosen to trust on their own. Over the evening, without meaning to, they took apart the AI-kills-SaaS argument. That argument runs like this: agents write code, Claude builds dashboards and ChatGPT generates interfaces in minutes, so once the software layer is a commodity, SaaS companies are in trouble. The inference sounds reasonable, but the conclusion rests on an assumption that never held: that SaaS was about selling software in the first place.
The stack nobody planned
None of the hosts were product managers or startup founders. There was a student earning money from her empty nights, a couple covering the cost of family visits and a local manager running other people’s properties. Another woman had built a small hospitality company without ever planning to. Their reasons for hosting had nothing in common, yet every one of them had the same structural problem.
Each of them runs a stack of tools around their listings. Dynamic pricing comes from PriceLabs or Beyond, guest messages fire automatically on booking status, and smart locks open on check-in windows. The listings run on Airbnb, Booking.com and VRBO in parallel, some with a direct booking site on top, and a few hosts already use AI for listing copy and review responses. None of these tools belongs to a single platform: each one is a third-party integration bolted onto a channel manager. That channel manager, Smoobu, is the one node in the network that every other tool has agreed to treat as the source of truth. It earned that place with reliable data, and the interface had little to do with it.
Reliability is the product
Nobody in the room said it out loud, but all of them understood it. What they pay for is a canonical record of bookings they can trust. The guarantee is easy to state: when Airbnb, Booking.com and the direct booking widget all take a reservation for the same Saturday night, exactly one guest shows up. It sounds mundane, and it is the entire product.
When that record slips, every tool in the stack fails in its own way. A pricing tool reading stale availability sells a night that was already taken. A messaging automation fires against a cancelled booking, and a revenue report pulls from one API while missing the transactions on another. Every automation in the stack shares one dependency, a central record it can trust, and when the record breaks, the stack breaks with it. It breaks at the worst possible moment, usually around 11pm and usually with an angry guest at the door. Strip away the interface and what Smoobu sells is data integrity as a service.
I’ve seen this position before
I spent three years at 3YOURMIND building software for industrial manufacturers such as Ford, the US Navy and SNCF. Their additive manufacturing operations ran across SAP and other ERP systems, machine APIs and the inevitable spreadsheets. Every attempt to build analytics, scheduling or cross-site coordination on top of those systems hit the same wall. Before anything else could work, someone had to answer one question: which system is right? The part of my work that mattered most was a reporting layer, one place where the data had been reconciled and could be trusted. Ford valued that layer more than the tooling it sat on, because it held data you could stake a real decision on.
| Inputs | One record of | |
|---|---|---|
| 3YOURMIND | SAP, ERP, machine APIs | manufacturing operations |
| Smoobu | Airbnb, Booking.com, VRBO | bookings |
A missing part in a 3D printing order and a missing booking have different consequences and one root cause. Something upstream was wrong, and every system downstream acted on it with confidence. The industries differ, while the company holding the reconciled record sits in the same structural position and sells the same product.
AI raises the stakes
The AI debate gets this exactly backwards. Its argument is that AI commoditises the interface: agents can call APIs, LLMs can generate dashboards, and the layer that used to be expensive is now nearly free. That part holds, and nobody should pay for an interface anymore. The hosts were never paying for one. They were paying for the data underneath, and AI makes that data more important.
AI is a multiplier, and it multiplies signal and noise alike. Clean data plus AI gives you better decisions at speed. Dirty data plus AI gives you worse decisions, faster and at scale. This is GIGO, garbage in, garbage out, with a bigger blast radius. The principle dates from the late 1950s, and it took nearly seventy years and autonomous agents for it to grow teeth.
Every new agent that reads your bookings, synthesises your pricing signals or messages your guests is one more consumer that breaks the moment the canonical record is wrong. Each layer of AI a host adds compounds the value of the data layer instead of diluting it. The interface has become nearly free, while the guarantee underneath it still takes years of work to earn.
The gap will widen
My bet for the next five years of SaaS is simple. Companies with clean, well-structured data at their core will get more out of AI than anyone else, because AI can act on what they have. Every model, agent or automation plugged into a reliable data layer performs. The same tools plugged into fragmented or stale data do the wrong thing with full confidence, like a cleaning team arriving to make up a bed for two when four guests are checking in.
The SaaS products that survive the AI transition will not be the ones with the best interface, because interfaces are becoming free. They will be the ones that have earned a place at the centre of the data graph. For the hosts in Berlin that place belongs to Smoobu, which thirty independent operators chose as their source of truth with no corporate mandate and no lock-in clause.
That choice rests on conflict-resolution logic, webhook retry handling and API versioning discipline. Behind it sits the work of mapping the quirks of Booking.com’s cancellation events, Airbnb’s availability model, VRBO’s pricing flags and every other platform’s idiosyncrasies into one coherent record. Then there is the monitoring layer that catches a sync failure at 2am before a guest does. None of this shows up in a demo, all of it took years to get right, and that is the moat.
Everyone in that Berlin office had independently made the same rational bet: the data layer is the last thing you want to rebuild from scratch. They’re right, and every agent they plug in from here makes the bet pay more, because it is one more tool acting on the record they chose.
We list our flat on Seumi Direct, built on Smoobu’s API. Direct bookings, no platform fee. It works because Smoobu’s availability data is reliable enough to build on.