Vibe coding, the bank teller effect, and the future of sync engines

In the 1960s, the ATM was invented. An ATM automates the core bank teller functions: depositing and dispensing cash. Bank tellers are/were humans serving other humans at a bank branch. At the time, everyone assumed bank tellers would be out of a job within a few years.

They were wrong. The number of U.S. bank tellers roughly doubled from 250,000 in 1970 to over 500,000 by the early 2000s. Why?

Go on, think about it. I'll wait.

FWIW I couldn't come up with any hypotheses either. Here's what happened: ATMs cut the cost of running a branch. Each branch needed fewer staff. But lower costs meant banks could open more branches with the same budget. More branches meant more tellers. The thing everyone was measuring - tellers per branch - went down. The thing that mattered - total tellers - went up.

The actual death of the teller came decades later with smartphones and online banking. The jobs were automated away eventually, just not by the technology everyone thought¹. I can recommend watching "The ATM Paradox: From Job Creator to Extinction Event" if you're interested in the history of all this.

Now let me tie this into app development. When agents hit the scene in 2025, my gut reaction was: this could foreshadow the death of apps. And by extension, the death of traditional app development. A few months later, however, a16z posted that app store submissions had actually gone through the roof.

This makes sense. AI-assisted dev has lowered the barrier to building, submitting, and maintaining apps. Exception: React Native is still life on hard mode :P but everything else got easier.

So we're in a boom right now, but I anticipate a bust. I expect app dev to follow the same dynamics as the bank teller story (60% confidence).

What could trigger a bust? I'm speculating here, but OpenClaw gives us a glimpse. Overhyped as it may be, it points toward an API-first world without apps. Your primary interaction with services happens through your personal assistant, not through tapping icons and form elements.

I already see this in my own behavior. Apps I barely touch anymore: Google, Reddit, Amazon, review sites. My agent handles most of what these used to do. I would love to be able to tell my PA: "here's my flight number, get me to the airport on time". Or, better yet, I set some budget constraints and my PA would just know I have an upcoming flight and ask me if I'd like an Uber for this trip.

What does this mean for sync engines? Near term, I anticipate healthy growth in adoption as a good chunk of the new apps in this cambrian explosion will need sync. Longer term (years) however, I expect sync engines will need to find new value props beyond "instant UI/UX". When the primary interface becomes chat, voice, video, etc, the instant UI/UX pitch will be less relevant.

Something we're well positioned to explore here at PowerSync is local agents:

As agents move on-device, they'll need their local state (did someone say SQLite?) synced with the always-on hiveminds in the datacenter. That includes pulling updates from other users and other edge agents, and only pulling the data they need (partial sync).

Agent-to-agent communication is a different story. Mostly they'll be online and can talk directly. Sure, "always online" systems can benefit from sync patterns: batching requests, recovering from failures. However, I suspect current day tools are already well positioned to handle these use cases.

Last but not least, offline will always be a significant chunk of use cases.