The claim that applications are disappearing is overstated in the way these claims always are. Nothing is disappearing. What is happening is narrower and more consequential: for a growing set of tasks, the interface a person touches is a conversation, and the application has become a service that something else calls on their behalf. That is not the end of software. It is the separation of the interface from the system, and it changes what you are actually buying when you buy an application.
When the interface moves to a conversation, the application stops being a product people use and becomes a capability something else calls. Very few vendors have priced for that
Here is what the shift actually means operationally.
What genuinely moves to conversation
Retrieval, first and most completely. Asking a question is a better interface than navigating to an answer, and this is already where most of the change has happened. Simple structured actions — submitting a request, updating a status, booking something. Anything where the underlying operation is one call and the hard part was remembering which screen it lived behind. Cross-system composition, which is the genuinely new capability. Questions that span four systems were previously a report request; they are now a sentence.
What does not move
Dense information comparison. A spreadsheet, a schedule, a dashboard — anything where the value is in seeing many things simultaneously — is worse as a conversation and will stay visual. Precision work, where the user needs to know exactly what will happen before it happens. Conversation is ambiguous by nature; configuration is not. And exploration, where the person does not yet know what they are looking for. Browsing is a real mode and language is a poor substitute for it.
| Task in the article | Interface to retain or test |
|---|---|
| Retrieval | Test conversational questions against access-controlled answers. |
| Simple structured action | Test exposed operations with confirmation and permissions. |
| Cross-system question | Test whether underlying systems can actually be called. |
| Dense comparison | Keep schedules, spreadsheets and other visual views. |
| Precision configuration | Keep explicit controls that show the proposed change. |
| Exploration | Keep browsing when users do not yet know the question. |
Qualitative summary of this article's source text, not a measured outcome or performance estimate.
The consequence most organisations have not priced
If the conversational layer becomes the interface, then application choice becomes an integration question rather than a usability one. The application with the better screens loses to the one with the better interface for machines to call, and that is a complete inversion of how software has been evaluated for thirty years. It also concentrates dependence. The conversational layer becomes the thing everyone uses, sitting above systems that are increasingly interchangeable, which is a strong position for whoever owns it and a weak one for everybody else.
What to do about it
Evaluate new systems on the quality of their machine interface as seriously as their screens. Ask whether the operations you need are exposed as operations, whether the permissions carry through, and whether a conversational layer could actually drive them. And keep the visual interfaces. The failure mode here is enthusiasm — organisations that push everything into conversation discover that the tasks people do daily are faster with a button.
Practical Guidance for Conversational Workplace Strategy
- Evaluate machine interfaces alongside user interfaces.
- Move retrieval to conversation first; it is the clearest gain.
- Keep dense comparison visual.
- Check that permissions carry through the conversational layer.
- Watch the concentration of dependence on one layer.
- Measure task completion, not conversation volume.
- Keep the direct interfaces available for daily work.
- Re-run application evaluations with integration weighted higher.
The Regional Angle
The first regional consideration is that conversational interfaces have an unusually strong case in a bilingual workforce, and an unusually uneven one. A person who works comfortably in Arabic but navigates an English-only enterprise system is much better served by asking a question than by finding a screen, and that is a genuine accessibility gain across a large part of the regional workforce. But the quality of that experience depends entirely on the model's Arabic competence and on whether the underlying system's data is intelligible when surfaced in translation, both of which should be tested rather than assumed. The second concerns the systems this layer would have to call, which in the Gulf mid-market are frequently heavily customised regional implementations. A conversational layer is only as good as the interfaces beneath it, and a system whose business logic lives in partner-built custom modules with no exposed operations cannot be driven conversationally regardless of how good the model is. For many regional organisations the honest sequence is to fix the integration surface first, which is a less exciting project than the one being proposed. The third is about where the conversation itself goes. Every question asked of a conversational layer is a record of internal activity, and if that layer is a hosted foreign service then the query stream is leaving the country even when the underlying systems do not. Regional organisations under residency expectations should treat the conversational layer as a data flow in its own right and assess it accordingly, because it is the part of the architecture most likely to be adopted without a review.
The objection worth taking seriously
The strongest objection is that this repeats a cycle the industry has run before. Natural language interfaces have been declared the future of computing roughly every decade since the 1980s, each time with a genuine technical advance behind them, and each time the graphical interface survived because precision, discoverability and speed matter more to daily users than expressiveness. Voice assistants were going to replace apps a decade ago; people still tap. Reorganising a software strategy around the assumption that this time is different is a bet against a very consistent historical pattern. That history is accurate and the scepticism is earned. The difference worth noting is what is being replaced. Previous attempts asked language to replace the interface for tasks the graphical interface did well — navigating, comparing, manipulating — and lost on the merits. What is actually working now is different: cross-system composition and retrieval, tasks that had no good graphical solution because they spanned systems no single interface covered. That is not a substitution, it is a new category, and it can succeed without anything disappearing. The right reading is not that applications are ending but that a layer has appeared above them, which is why the practical advice is about integration quality rather than about retiring interfaces.
Common Questions
Should we stop buying applications with good interfaces?
No. Weight the machine interface higher than you used to, and keep the screens, because daily users still prefer them for repetitive work.
What breaks most often in conversational deployments?
Permissions. The layer frequently runs with broader access than the asking user has, and the failure is invisible until someone gets an answer they should not have.
Is this cheaper than building interfaces?
Rarely, at first. The saving comes from not building interfaces for the cross-system cases, which you previously handled with reports and requests.
What should we expect over the next twelve months?
Expect conversational access to be bundled into platforms rather than bought separately. Expect integration quality to appear explicitly in procurement scoring. Expect a visible failure involving permissions in a conversational layer. And expect visual interfaces to remain exactly where they are for daily operational work.
Conversational Workplace Strategy — we start with the integration surface, because the conversation is only ever as good as what it can call.