I have been thinking about how AI & agents are impacting the software/app business and I realized a major shift is imminent - in fact, it is already underway: from apps for users toward apps for AIs.
There is no question that humans will continue using apps on their devices, including ones tailored for specific needs and scenarios. However, the main app, the main "entry point" will increasingly be their AI assistant. It is happening already - people are turning to Claude, ChatGPT, or Gemini with all kinds of things, also leveraging the voice interface (which AIs have finally made usable) rather than turning to specialized apps and sites. I think in the long run - within a decade or so - UI will be a tailored experience that will be generated on the fly by AI assistants for each user, adapting to what the user wants to do at the moment, from playing a game to reviewing astrophysics research. As AI assistants become increasingly the main gateway for their users (just as smartphones and tablets have displaced the PC over the past decade or so - hence the recent focus on web apps/SaaS as opposed to desktop apps), AIs themselves will become the main users of all other systems. This is so because AIs need current data and specialized tools to function outside of being a generic chat.
This shift is already happening. MCP (Model Context Protocol) facilitates exposing data and systems performing specialized functions to generic LLMs and frameworks that support it - and it is fast becoming the standard (Anthropic's Claude can use MCP directly, agentic frameworks like AG2 wrap MCP servers as tools, even OpenAI includes MCP support now). So, if you expose your service or data via an MCP server, you will soon get AI users utilizing it - while their human users will just pay you for their usage.
Let's look at an example that is a reality already today. Say you want to analyze your investments. You purchase access to an analytics system that makes its functionality available via the MCP, configure the connection (which in Claude now is as simple as providing a link to the MCP server and going through OAuth authentication), and suddenly your assistant can not only answer general questions but actually analyze data, test strategies, and create recommendations. You still pay for access to the analytics system, but your AI agent is the one using it. The analytics system might not even have a UI for human users - except of course for handling payments and authentication - and provide all of its specialized functionality exclusively through MCP.
But this is just the first layer, first step. The next is agents being exposed as services for use not by humans, but by other agents. This is what the A2A protocol proposed by Google is for.
Imagine this scenario: your personal AI assistant needs to evaluate a potential business acquisition. Instead of trying to do it itself (which might not be its strong suit), it contacts a specialized market analysis agent. That agent, in turn, might consult a financial modeling agent for valuation, a competitive intelligence agent for market positioning, and a technical due diligence agent to assess the target's technology stack. Everything happens automatically in the background, you receive a comprehensive acquisition report - and a recommendation. It is only logical to expect that very soon even the payment for services of those other agents and tools will be done automatically by your assistant, without you having to go (again) through registration and providing payment details on yet another platform.
So, my prediction is that the software economy will change within a few years from human-centric to AI-centric. Yes, people will be using apps - as I mentioned above - but AIs will be the main users of tools (through MCP) and agents (through A2A or a similar protocol) to achieve specialized functions.
If you are a software vendor, you should rethink your strategy now. Maybe instead of wondering how to make your system's UX better, think about how to make its AX - AI Experience - better.