Every few years someone declares the graphical user interface (GUI) dead. Chat will replace everything. Voice will replace chat. Then the excitement fades and we all go back to our spreadsheets.
Reality is less dramatic. New kinds of interfaces don't replace the old ones. They get added alongside them.
What GUIs still do best
GUIs win when users need to see many things at once: prioritization boards, analytics dashboards, design tools, maps. Users rely on spatial memory: they remember that a certain chart sits in the bottom-right corner. They also rely on direct manipulation: dragging, resizing, and selecting several items at once. Trying to apply layers of filters through chat alone still frustrates experienced users.
Most professional software is still built around a GUI for a good reason: the work involves comparing things visually, and conversation makes comparison slower.
What conversational interfaces do best
Conversational user interfaces (CUIs), whether text or voice, win when you know what you want but not exactly how to get it. "What's wrong with this contract clause?" "Summarize these three listings." "Walk me through the setup." In each case, you know the outcome you want before you know which menu to open.
They also work well on mobile, where the screen is small and tapping through nested settings menus is painful. But this only holds if the interface responds quickly and recovers gracefully when it misunderstands you. That is a higher standard than polished demos make it look.
Hybrid is the unglamorous answer that usually works
The products that feel ahead right now mix both modes: a GUI holds the structure and the saved work, and conversation speeds you up within a single step. Think of an assistant built into a form field, rather than a blank chat window that forgets what you built yesterday.
As a product leader, I ask one question for each workflow: what is the user actually working on? If it's a dataset, a timeline, or a canvas, keep the interface graphical. If it's a question or a story, conversation can lead. If it's both, embed a conversational assistant while keeping the underlying data visible.
Consider analytics: a CEO asking "why did churn spike in March?" might start in chat, because natural language is a fine way to frame the question. But answering it well still requires charts, customer segment breakdowns, and filters. The conversation should open the GUI. The same goes for design tools, CRMs, and M&A due diligence workspaces: the lasting record lives in the structured interface, and the assistant is a shortcut into it.
Voice has its own strengths: it helps when your hands are busy, it improves accessibility, and it lowers the barrier for people who struggle with reading or typing. But a voice interface that acts without confirming and showing the result makes users nervous when money or reputation is on the line. Read the action back. Show the record. Let them tap to edit. GUI habits exist because people verify things with their eyes.
Design implications for founders
Map out your users' tasks before you rewrite your product as one long chat thread just because investors are excited about agents. For each task, choose the interface based on three things: how costly a mistake would be, how much comparison the task requires, and how often people do it. Frequent, precision-heavy tasks deserve graphical controls. Occasional guidance can happen in conversation.
Document the choice. Otherwise, future you (and future hires) will default to "add a chatbot" because it's fashionable. A one-page interface map for each major workflow saves months of rework later.
The real interface question is which tool fits each task. That is a deliberate choice, not one mode applied everywhere.
Originally published on Product, AI & Business on LinkedIn.