The Friction Economy: Why AI is Killing Trust
If the 'cost' of generating text is no longer a proxy for how much someone actually cares, we have to rethink how we build trust in our communications.

Automation needs a narrow first win
The best first AI workflow is usually a repeated task with a clear input, clear output, and a human approval step.
LLMs have effectively commoditized the act of putting thoughts into writing. When anyone can generate a coherent, well-structured message in seconds, the perceived effort behind that message drops to zero. We’re hitting a wall where the ease of production makes it nearly impossible to prove genuine commitment. If the 'cost' of generating text is no longer a proxy for how much someone actually cares, we have to rethink how we build trust in our communications.
The Return of High-Friction Media
To counter this, we’re seeing a massive shift toward methods that require physical effort or high-bandwidth interaction. This isn't just some 'retro' trend; it's a move toward 'proof of care.' Handwriting and physical flyers are resurfacing because they represent a tangible investment of time and physical movement. It’s about intentionality. Even as technology tries to bridge this gap—like people using pen plotters to create imitation handwritten flyers at scale—the underlying goal remains the same: finding a way to signal that a message wasn't just copy-pasted from a prompt.
We’re also seeing a move toward things that are harder to fake with an LLM. Tattoos and public storytelling are becoming the new gold standard for broadcasting commitment. They involve permanent physical changes or real-time, high-bandwidth performances. When you can't easily automate a face-to-face gathering or a live performance, these interactions become the primary way to transmit complex ideas that require a high degree of trust.
Engineering Signal in a Zero-Cost World
What this actually points to is the erosion of signal in our digital interactions. As builders, we’re trained to optimize for efficiency, but we’re reaching a point where efficiency has destroyed the value of the output. If a message costs nothing to produce, it conveys nothing about the sender's intent. The real story here isn't that the technology is failing to write well; it’s that it’s succeeding so well that it has stripped away our ability to distinguish between a human thought and a machine prompt.
As we integrate more AI into our workflows, the real question isn't whether the demo works. It’s what happens when every piece of communication is perfectly polished but carries zero weight. We’re going to see a massive premium placed on high-bandwidth, 'high-friction' interactions because they are the only remaining ways to verify that a human is actually in the loop. If you want to build trust in a world of automated content, you have to find ways to reintroduce the friction that AI has removed.


Got a question about how this applies to you? →
Keep reading
Follow the thread
The Technical Debt of Unconsented Training Data
If we’re building tools on a foundation of stolen labor, how long before the 'technical debt' of unconsented data crashes the whole system?
Read this noteSame lane, different angle
The Linux Kernel's AI Pivot: Technical Merit vs. The Verification Bottleneck
If we're strictly chasing technical excellence, what happens when the 'best' code is just the code that aligns with a corporate owner's hidden priorities?
The Robustness-Generation Gap: Why Dialect Adaptation Isn't Just a Fine-Tuning Problem
Most researchers think high benchmark scores mean a model has mastered a dialect. DiaLLM research shows that's not true: understanding a dialect and actually speaking it are governed by two different parts of the training pipeline.