Surviving AI: Designing what machines can't read

We forget most of what we experience. What survives is meaning, not pure information, and is uniquely human.

Coding agents have condensed software production. As more knowledge workers consolidate context and invite agents into their work, the horizon when everything that can be accurately translated via text is automated gets closer.

The conviction that comes through struggle, the instincts that are developed through repeated failure, the natural delete button that continuously cleanses our own biological context window: these things can't be stored as markdown.

AI has changed how we design in obvious ways, but its parallel impact on how users perceive technology is both more subtle and more profound. Machines can't fully read it, pre-trained models can't predict it, and no one fully understands it. But we all feel it.

The lasting design problems are increasingly human. Trust, transparency, and feel are new heuristics. AI can optimize based on infinite data points, but it takes a human to decide what not to optimize that will make a product truly feel great.