# AI Is Becoming the UI: What It Means for Users and Software Companies By Dan Seyer, Founder, Build Marketing · Published 2026-10-30 > AI is changing how people use software. Instead of opening an app and clicking through menus, more of us describe what we need and let an assistant do the work across the systems it's connected to. For users, that means less friction, fewer repetitive tasks and access to things that used to need a specialist. For the companies that make software, the questions are harder. If customers work through someone else's AI, what makes your product valuable? How do customers find you? And what can you charge for? This is part 2 of The Enterprise Platform Battle. [Part 1](https://www.buildmarketing.ai/the-new-enterprise-platform-battle) asked who owns the platform. This one looks at the same shift from both sides, users and product companies, and at where the old screens still matter. ## AI Summary _Summary by AI._ - AI is becoming the interface to software: people describe what they need, and an assistant works across the systems it is connected to. - The phrase "AI is the new UI" dates back to Accenture in 2017. MCP, released in November 2024, made it practical, and ServiceNow, Zoho and NetSuite now say it openly. - For users, the article lists less friction, fewer repetitive tasks and broader access, limited by what the AI can reach, what it is permitted to do, and the need to check its work. - For product companies, the value moves from the screen to functions, data and business rules, and discovery and pricing change with it. - Screens remain for review, approvals, design and setup, but increasingly the AI generates them on demand. ## An Old Phrase With New Plumbing "AI is the new UI" isn't a new line. Accenture made it the first trend of its Technology Vision in January 2017, almost six years before ChatGPT. 79% of the executives it surveyed agreed that AI would change how they get information from customers and interact with them. The idea was right. The technology wasn't ready. In 2017 an assistant could set a timer. It couldn't close your books. What changed is the plumbing. In November 2024 Anthropic released MCP, the Model Context Protocol. Wim Sweldens described it well in his July essay "AI is the new UI": "an open standard that lets an AI system see what a piece of software can do and call it directly." Once that exists, he argues, "the very idea of an app having its own user interface is about to become optional, and then obsolete." Then the software companies started saying it too. - ServiceNow launched AI Experience in September 2025 under the line "AI is the new UI," and called it "a single, intuitive interface" for work across its platform. - Vala Afshar of Salesforce and Henry King wrote "AI is the new UI" for CIO in October 2025. Their point: "It's not a new screen; it's your enterprise relationship layer." - Sridhar Vembu, Zoho's co-founder, put it most plainly in February. Agents would "contextually integrate the underlying SaaS apps and offer a vastly richer and easier user experience. In other words 'AI is the UI' for SaaS." In September Zoho shipped Zia Chat, which lets you talk to all your Zoho data and take actions. - NetSuite said it at SuiteWorld 2026 in Las Vegas this week. Ask Oracle is now the front door to NetSuite, not a chatbot sitting beside it. And more than 13,000 NetSuite customers already use AI through its AI Connector Service over MCP, with Claude or ChatGPT. Vembu's version is the one I agree with most. AI becomes the layer you talk to. The applications underneath supply the capabilities. ## What It Means for Users We see this every day, for ourselves and for our clients. Three things change. Less friction. You don't need to remember which app has the answer or where the report lives. You ask. A question that used to take three logins and an export takes one sentence. Fewer repetitive tasks. The copy and paste between systems is the first thing to go. So is the weekly report someone builds by hand. Broader access. This is the one I care about most. A lot of what enterprise software can do never gets used, because only a few people know how. The report builder. The workflow rules. The advanced search. When you can ask in plain English, everyone on the team gets the power user's tools. There's a catch. Your AI is only as good as what it can reach and what it's allowed to do. If a system has no connector, it's invisible. If permissions are wrong, the AI either can't help or helps too much. And you still have to check the work. An answer that looks finished isn't the same as an answer that's right. This prompt shows which of your apps you could already run from your AI tool. **Prompt: Which of my apps could I run from AI?** (for your favorite AI tool (ChatGPT, Claude, Gemini, Copilot)) Lists the software you use most, the jobs you do in each, which of those you could just ask your AI tool for, and what you'd still open the app for. ```text I want to find out which of the software I use every day I could run from my AI tool instead of opening each app. First, ask me one question at a time, and wait for my answer before asking the next: 1. Which AI tool do I use most, and which connectors or integrations do I already have turned on? 2. What are the ten apps I spend the most time in at work? 3. For each app, what are the three jobs I do in it most often (for example pull a report, update a record, approve something, send a campaign)? 4. Which of those jobs do I do only because I know where the button is, and which take real judgment? Then give me: - A table with one row per job: the app, whether that app has an MCP server, connector or API my AI tool can use (say if you aren't sure), whether I could just ask for it in plain English, and what could go wrong if the AI did it. - The three jobs I should try from my AI tool first, with the exact sentence I'd type for each. - The jobs where I should still open the app, and why (for example reviewing a lot at once, approvals, design, setup). - What I'd need from IT or the vendor to make the rest possible, such as a connector, permissions or a security review. Keep the language plain and short. Don't guess about connectors you can't confirm. Mark them as "check with the vendor". ``` More prompts: [Prompt Library](https://www.buildmarketing.ai/prompts.md) ## What It Means for Product Companies This is the harder side. First, it helps to see how it works under the hood. ![How a request flows when AI is the UI. The user types in chat. A request router decides. A recognized command goes straight to a prewritten function. A request that needs language interpretation goes to AI, which interprets it and calls the same prewritten function. The function uses the product's data and services, and the result is shown in chat or a visual interface.](https://www.buildmarketing.ai/images/ai-is-becoming-the-ui-request-router.svg) The user types in chat. A router decides what to do with it. If it's a command the product already knows, it goes straight to a prewritten function. No AI needed, which is faster and cheaper. If it needs language interpretation, AI works out what the user wants and then calls the same prewritten function. Either way, the function does the real work against the product's data and services, and the result comes back in the chat or in a visual interface. Look at what's in the middle of that picture. It's not the screen. It's the functions and the data. That leads to three questions. What makes you valuable? If the user never sees your screen, you can't win on the screen. You win on what's underneath: your data, your business rules, your permissions, your audit trail. In Part 1 I said headless doesn't mean worthless. That's still true. But you have to be the system the AI trusts to get the work right. Sweldens describes a future of apps "competing purely on backend capability." Many product teams spent the last ten years competing on design. That's a big change. How do customers find you? When an agent picks the tool, the buyer may never see your brand. Discovery moves to connector directories, to how well your functions are described to machines, and to whether an agent can read your website and docs at all. That's why we built [MachineReady](https://www.buildmarketing.ai/machineready/), to see how AI reads a site. I wrote more about it in [Does Your Website Need to Speak Agent?](https://www.buildmarketing.ai/does-your-website-need-to-speak-agent) What can you charge for? Seats paid for people looking at screens. If people stop looking, seats stop making sense. I wrote about the meters in [SaaS Is Bringing the Work to You. Then It Sends the Bill.](https://www.buildmarketing.ai/saas-is-bringing-the-work-to-you) The vendors that learn to charge for work done, not for logins, will keep their customers. The ones that don't will watch usage go up and revenue go down. There's also a fork in the road. Do you bring your product into Claude and ChatGPT, or build your own assistant and ask users to come to you? NetSuite is doing both: Ask Oracle inside NetSuite, and MCP for the customers who would rather work in their own AI. I expect most vendors to end up doing both. This prompt checks whether your product is ready for that. **Prompt: Is your product ready for AI as the UI?** (for your favorite AI tool (ChatGPT, Claude, Gemini, Copilot)) For product teams: maps your top customer jobs to the functions an AI assistant would call, the data they need, how customers would find you inside someone else's AI, and what you'd charge for. ```text I lead or work on a software product. I want to know how ready it is for customers who never open our screens and work through an AI assistant instead, like Claude or ChatGPT. First, ask me one question at a time, and wait for my answer before asking the next: 1. What does our product do, and who uses it? 2. What are the ten jobs customers do in it most often? 3. Do we have a public API, an MCP server or a connector in any AI tool today? What can it do? 4. How do we charge today (seats, usage, tiers, credits)? 5. What makes customers choose us over competitors today? Then give me: - A table with one row per customer job: the function an AI assistant would need to call, whether we have it today (API, MCP or neither), the data and permissions it needs, and whether it should be a fixed command or needs AI to interpret the request. - The jobs that still need a screen, and whether that screen could be shown inside the AI assistant instead. - What still sets us apart if nobody sees our interface: data, business rules, permissions, audit trail, integrations or something else. Be honest if the answer is mostly design. - How a customer would discover us from inside someone else's AI: connector directories, how our functions are described, and whether an agent can read our website and docs. - Two or three pricing options that charge for work done instead of logins, and the risk in each. - The first three things to build, in order. Keep the language plain and short. ``` More prompts: [Prompt Library](https://www.buildmarketing.ai/prompts.md) ## Where the Screens Still Matter I don't think screens go away. Some work needs them. - Reviewing a lot at once. A close checklist, a pipeline, a campaign calendar. You want to scan it, not read paragraphs about it. - Approvals. When money moves or a contract goes out, people want to see exactly what they're signing off on. - Design and creative work. You have to see it. - Setup and admin. Someone still configures the system the AI talks to. What changes is who builds the screen. More and more, the AI builds it when you need it. NetSuite's new AI Portlets build a dashboard from a question. As Craig Sullivan, who leads product management at NetSuite, put it, "You literally just have to ask it what you want to see." And MCP apps show a vendor's own screens right inside the AI assistant. The UI doesn't disappear. It shows up when you need it and goes away when you don't. ## What I'd Do Now If you buy software, ask each vendor three things. How does your product work from my AI tool? What does it cost when it does? What do I lose if I never open your app? If you build software, try using your product with no screen at all. Give an AI assistant your API or MCP server and your top ten customer jobs, and see what breaks. That's your roadmap. **The Enterprise Platform Battle, part 2 of 2:** [AI becomes the interface](https://www.buildmarketing.ai/the-new-enterprise-platform-battle.md) · AI is becoming the UI If you found this article interesting and want to talk about AI and your organization, [reach out](https://www.buildmarketing.ai/#Get-In-Touch). _Sources: Accenture, "Accenture Technology Vision 2017 Forecasts a Future of Technology for People, by People," January 26, 2017. Anthropic, "Introducing the Model Context Protocol," November 25, 2024. Wim Sweldens, "AI is the new UI," AIQ, July 24, 2026. ServiceNow, "ServiceNow Unveils AI Experience, the UI for Enterprise AI," September 30, 2025. Vala Afshar and Henry King, "AI is the new UI," CIO, October 23, 2025. Sridhar Vembu on X, February 22, 2026, and September 23, 2026 (Zia Chat). Diginomica, "SuiteWorld 26: NetSuite takes the wraps off new agents and skills," October 26, 2026 (13,000 customers on the AI Connector Service, AI Portlets, Craig Sullivan quote). Australian FinTech, "NetSuite shows the mechanics of the autonomous suite in SuiteWorld 2026 product keynote," October 29, 2026. TechRadar, NetSuite MCP apps at SuiteConnect London, March 31, 2026._