There’s a shift coming, as the Chatbot Age enters its waning months and everyone is trying to anticipate what the next big step for LLM software is.

Agentic AI is one path forward, a promise to hand off the many tasks and errands of your life—handling communication with co-workers, completing a project between multiple pieces of software, buying groceries—to an AI. Or in other words, give your credit card number, logins, all other personal data and a post-it note to hallucinating bots that respond with “you’re absolutely right” to all your absolutely wrong requests. How could it possibly go wrong?

Another anticipates an end to the old ways of digital interaction, where software companies craft apps that anticipate the needs or desires of their users. They design a visual interface, build the logic behind that interface that communicates with the many necessary systems, and deliver an output that makes you happy (or addicted). Launch that software, get a userbase, and charge people (or stripmine them for data). Improve and iterate.

There are problems with all that, no doubt, but it all makes good enough sense.

The new way promises to be done with much of that. It promises an ever-shifting, amoebic middle between a user’s desires and the eventual output. No more rigid paths or complex user interfaces. No more purpose-built software with clearly defined logic. It’s just: “make request; get result”. Or, if that isn’t quite right, refine until it is. It’s DWIM—Do What I Mean. It’s GenUI—custom interfaces built on-the-fly, just for you. It’s Tom Cruise in Minority Report. We’ll communicate with the data-driven world through a highly-personalized layer of software, where AI bots are the proxy for every request. They’ll predict our intentions and morph interfaces to match them. My experience and my friend’s experience with any given piece of software might be totally different.

Let’s set aside the security vulnerabilities of having AI agents—sycophantic to any request anyone makes no matter how ridiculous or absurd—be the proxy for every interaction between data and user. Let’s set aside the insane compute requirements to reform UIs on the fly and handle the vast majority of logic pathways. Let’s set aside what developing inside such an ecosystem might look like—are we all just writing structured JSON for bots to consume? Or LLM.txt files? Or MCP clients?

Let’s set aside all that.

I worry about what total surrender to convenience looks like; the loss of friction. A world of hyper-personalized software might be what we think we want. It might even be what we want. But what do we lose in a frictionless digital world?

We lose the joy and wisdom of the journey.

The strange, unique, intimate corners of the web become ghost towns. The deeper understanding that comes only to those who spend time browsing, reading, looking, learning, researching slips away. Scrolling through a grid of every photo you’ve ever taken isn’t the most efficient way to find the one you’re looking for. But without it, how do you get ambushed by that lost image, triggering the sights, smells, sounds, and feelings of that long-forgotten night?

Lately, I’ve been trying to use the internet and apps in my life similar to the way I did when Netscape Navigator first booted up on my parent’s Windows 2000 PC. The wonder of chatting to multiple friends on MSN. The near addict behaviour of one more StumbleUpon click. Spending hours drawing individual GIF frames in Macromedia Fireworks.

I think we lose a lot in a frictionless digital world, one where fawning AIs try ceaselessly to appease our every whim, every discomfort, every errant desire. I want to enjoy the journey.