Everyone and their cat is trying to add AI to their apps, SaaS and businesses. Usually it's some sort of chat or sparkly "AI" button that's placed at the most visible spot on your site or app. It often looks very annoying, works poorly, users hate it - but the companies still push it, because if they aren't "AI-native" in 2026, they're pretty much doomed.
And it makes sense it works poorly. If the company uses a powerful model, it will cost A LOT. So, either the price for the service goes significantly up to justify increased costs, or the company has to use cheaper models to keep the price down.
But users likely already pay for a ChatGPT or Claude subscription. And have access to the most powerful models. And got used to the chat interface there, and have all the context and memory. They feel way more comfy being there. So, IMHO the right way is not to integrate AI into your app/site, but to let users connect to your service and use it from their ChatGPT - i.e. good old API!
There's a standard for that - Model Context Protocol (MCP), which frankly is a bit ugly and overcomplicated, but all the major AI-providers (Claude/ChatGPT/Gemini) support it, and it handles auth. It basically gives the agent the commands (like API endpoints), and instructions on how to use them.
If you don't have MCP for your SaaS, app, website, business - you're missing out. That's IMHO the right way of adding "AI" to your business. Not adding another AI chat client to your app, or adding this sparkles button that will apply "AI" to whatever your app's output is. But letting Claude/ChatGPT/Gemini/etc do the things that users had to do previously through your UI. Because before users couldn't use API directly. Now they kinda can.
As developers, we already use that heavily. For example, I have MCPs connected to the iOS simulator, devtools, Google Ads, analytics, error aggregators - and I'm sitting like a spider in the center of my web, getting data from all those things, and telling the agents to do stuff with it.
And non-software-eng-people are also catching up. They want to delegate booking their flight tickets, hotels and museums. They want to search for homes to buy or rent, book tables in restaurants, update their credit card on insurance or utility sites, etc etc. Because all that work is annoying and takes time to do. So, provide an API/MCP for that! Otherwise, your competitors will.
I've got a small business - a weightlifting app. Almost every single thing there could be configured through MCP. You can make the agent set up available equipment, enter all the plates, create workouts, create lifting programs, validate them, write the logic for them. And it's amazing what people do with it.
Agents write the lifting programs for users, help to visualize the progress, and recommend the next steps. The app has graphs, but Claude can build way richer personalized visualization, HTML/CSS/JS right on the spot.
My favorite example is when a user told me they just went around the gym, took pictures of all the equipment pieces - machines, plates, dumbbells - uploaded the pics to ChatGPT, and told it to update the equipment settings in the app according to the pics. And it worked perfectly.
There's a concern though that if the AI agent can do anything the app can do - why do we need the apps then? Will they die eventually? Your business survives because it owns the data and does things LLMs can't do on their own (available houses, tickets, venues, etc). Your UI survives wherever it's faster than chat - like tapping a button on a watch to log a set vs typing in the chat what you just lifted. The parts of the UI that are just forms over an API though - those may die. But that's ok.
The funny thing is - the "AI-native or doomed" panic is justified. If you don't integrate into the AI wave somehow in 2026, you may be doomed. It's just not about having an AI button on your site - it's about giving the AI ecosystem convenient access to your site, app and data. Not the sparkles button.