Are We Building Tomorrow's Legacy Code?
Here's an uncomfortable question I've been sitting with lately.
We're all racing to build AI-powered web apps. Chatbots, document processors, code assistants, productivity tools. The market is exploding, and everyone wants a piece. But what if we're building on the wrong layer?
The Shift Happening Right Now
MCP (Model Context Protocol) is changing the game. Instead of users opening your web app, logging in, and using your interface, they can now connect directly to AI agents that access your functionality through MCP servers.
Think about what that means:
- No more context switching between apps
- No more copying and pasting between tools
- No more "let me open that other tab"
The AI agent becomes the interface. Your carefully designed UI? Your onboarding flow? Your sticky features that keep users coming back? They might all become optional.
We've already seen glimpses of this with ChatGPT Apps and custom GPTs. Users don't visit separate websites anymore. They just talk to their AI agent, and it handles the rest.
The best interface is no interface by Golden Krishna
A Pattern We've Seen Before
Remember when every business needed a mobile app? Newspapers gave way to radio. Radio to TV. TV to internet news. Desktop sites to mobile apps. Mobile apps to progressive web apps.
Each shift felt revolutionary. Each time, the old guard insisted their medium was irreplaceable. Newspapers said radio couldn't deliver real journalism. Radio said TV was a gimmick. Desktop advocates said mobile was too limited. They were all eventually proven wrong.
Are we watching it happen again? Building browser-based AI tools while the infrastructure shifts underneath us?
What Actually Survives
Here's the nuance everyone's missing: MCP doesn't kill all web apps. It kills bad web apps. The thin wrappers around AI capabilities. The ones that are just a pretty chat interface with some prompts. The ones that don't provide real value beyond access. If your product is genuinely solving a complex problem, creating unique workflows, or building network effects, you're not just safe. You're better positioned. Now you can reach users through multiple channels: your web app, your API, and through MCP servers that AI agents connect to.
The question is: which category are you in?
What This Means for Builders
Your core value needs to run deeper than "we made AI easy to use." That's table stakes now. And it's getting easier every quarter as the models improve. Here's what I'm doing differently:
Building MCP servers alongside web apps. Not as an afterthought. From day one. If I'm building something users should access through an AI agent, I'm making that path as smooth as the web interface.
Focusing on workflows that are hard to replicate. Simple Q&A? That's commodity. Complex multi-step processes with state management and business logic? That's defensible.
Accepting that distribution is changing. The metric isn't just "monthly active users visiting our site." It's "services connected through our infrastructure." That's a mental shift, but it's necessary.
The Reality Check
Models improve every quarter. What requires a custom interface today might work through a simple conversation in six months.
I've seen this firsthand. Features I built elaborate UIs for six months ago? Users now just describe what they want in plain English, and the AI figures it out. If your entire moat is UI/UX around AI capabilities, that moat is evaporating. Ask yourself honestly: "If users could access this through their AI agent without ever opening my app, would they?"
If the answer is yes, you might be building temporary infrastructure.
The Uncomfortable Truth
I'm not saying stop building web apps. I still build them. But I'm asking harder questions about what layer I'm building on.
Am I creating lasting value, or am I just filling time until the next platform shift? Am I building the Dropbox of AI tools (essential infrastructure), or am I building another note-taking app that'll be replaced by a native feature?
The winners in this shift won't be the ones with the prettiest AI chat interface. They'll be the ones that become essential infrastructure that agents need to connect to.
What's Next
We're moving past "there's an app for that." Soon, apps and protocols might both become irrelevant. AI will just understand what you want and make it happen, connecting to whatever services it needs behind the scenes.
Your job as a builder? Make sure your service is one it needs to connect to. Not because you have the best landing page or the smoothest onboarding flow. But because you solve a real problem that can't be replicated by a smarter model and a better prompt.
That's the bet I'm making. And honestly? I'm not sure if I'm early or late. But I'd rather ask these questions now than in 18 months when the answer is obvious.
Are you ready?