Woodson Martin, chief executive of the enterprise software company OutSystems, uses Google Maps every day. He uses it to find a place to order pizza, to get directions while driving and to walk around foreign cities he doesn't know well. That habit became his test case for a big question in software: will the apps we use now dissolve into conversations with AI assistants?
His answer, given on The Cognitive Revolution in an episode published October 7, 2026, was a qualified no. "I personally still imagine that a year from now I'm going to have more than a hundred apps on my phone," Martin said. Some of today's software could still break into smaller pieces, he added. OutSystems is one of the episode's sponsors.
Two ways to use software with AI
The host, Nathan Labenz, described two patterns. The first is the AI sidebar. OutSystems' version is called Mentor, and Labenz described it as wrapping the existing application. Tasks that users once did by dragging and dropping elements on screen can now be carried out by the AI through tool use, meaning the model calls the application's functions directly. Users can still do things the old way, so both options exist side by side.
The second pattern comes from MCP, the Model Context Protocol. MCP is a standard that lets AI assistants connect to outside tools and data. With MCP, Labenz said, a user might never open the original platform. They could work from some other piece of software, or "take a walk and just use voice mode" to talk to an agent. In that case the platform's tools might never appear on screen in the form they take inside the platform itself. He asked Martin where OutSystems expected things to go.
The case for widgets, and its limits
Martin said not every experience suits a conversation, whether spoken or typed. Even so, he saw "a real significant chance that a lot of software as we know it today is just widgetized into little things that show up in the flow of conversation."
Google Maps showed him where the conversational approach falls short. He said he could already ask ChatGPT for turn-by-turn directions. But he would still want a map: "what the hell does the world look like around me? How is it shaped? What am I going to pass on the way?" He did not think conversation would handle that perfectly. Privacy, compliance and plain user preference were all reasons, he said, to expect "a wide variety" of ways to use software in the future.
Martin did accept that the large, all-in-one systems common today could change. "A portal I go to with 80 features" might instead be delivered "in little chunks, just when I need them in the flow of other work," he said.
The standards needed for that kind of chunked delivery have started to appear. In January 2026, the MCP maintainers announced MCP Apps, an extension that lets a tool return an interactive interface inside a conversation. Their examples included a sales dashboard the user can filter, a deployment form and a contract with highlighted clauses to review. Choices the user makes in these interfaces can update what the model knows without the user typing a new prompt. For security, the interface runs in a sandboxed frame with restricted permissions, and hosts can require explicit approval for tool calls the interface starts.
From a remodeling chat to a loan application
To show what this could look like in practice, Martin described a demonstration from OutSystems' user conference in June. In it, one of the company's banking customers started loan applications from a consumer's conversation with Claude, Anthropic's AI assistant.
In the demo, the consumer is talking with Claude about remodeling a home and thinking about a new stove. The cost starts to add up, and the consumer says they might need to borrow money. They are told they are pre-approved for a $30,000 loan and asked whether they want to start an application. The process then moves from Claude into the bank's mobile app. According to Martin, the same agent keeps working in the background in both places and carries the conversation over.
Because the consumer is already a bank customer, the system recognizes which standard loan documents the bank already has. "They don't need to ask you for your bank statement," Martin said. The application then goes through the bank's back-end systems. There, agents either handle the work or make recommendations to humans who review each stage.
OutSystems announced its first end-to-end banking product for loan origination, the process of taking and approving loan applications, on June 2, 2026. At the time, it said the product would be available in early Q3. The company's product page describes its industry solutions as pre-built agentic systems that come with an audit trail, with industry-required approvals built in, and says customers can start with a single component. Martin did not say whether the conversational demo used that product.
Why the bank app stays
Martin said the demo shows both sides of the question. Much of the work moves into the agent conversation. But the mobile app remains because the bank needs it to meet compliance, privacy and regulatory requirements. Asked whether those requirements might eventually change so the app is no longer needed, Martin said, "I don't know. We'll see."
He said the question that interested him most was not whether all software would end up looking the same. It was "what will the imagination now of creators produce now that AI and platforms like OutSystems remove so many of the typical traditional constraints of building amazing experiences?"