Calm software is a feature

The software I admire most has a quality that never appears on a feature list: it's calm. It does its job and then gets out of the way. In an attention economy, that restraint has become a radical act — and a competitive advantage.

Attention is the real cost

We price software in euros per month, but users pay in attention: notifications answered, badges cleared, interfaces re-learned, decisions demanded. Most products are attention-expensive because engagement is what their business measures. The tragedy is that the pattern spread to tools with no engagement motive at all — internal systems, booking tools, education platforms — through pure imitation.

When I build for small businesses, the brief behind the brief is always the same: make the computer part smaller. The cleaning company doesn't want a dashboard experience. They want to know which sites are done. The difference between those two sentences is the entire discipline of calm software.

Principles of calm software

The working rules I've converged on, project after project:

• Default silence. Speak only when the user must act or genuinely wants to know. Every notification is a withdrawal from a finite trust account.
• One screen, one question. If a screen asks three things, it's three screens wearing a trench coat.
• Familiar first. Model the workflow people already have — the paper diary, the shift note — and innovate only where the old way actually failed. Novelty is a cost, not a feature.
• Fast is calm. Latency is a tiny alarm bell. A system that responds instantly feels like a tool; one that lags feels like a negotiation.
• Finish states. Calm software lets you be done. No infinite feed, no red badge regenerating overnight, no guilt engine.

None of this requires genius. It requires refusing, repeatedly, to add the thing everyone else adds.

Calm and AI: a warning

AI can be the calmest technology ever shipped or the loudest, and the industry is currently choosing loud: assistants that interject, summarise the unrequested, and gamify their own usage.

The calm version is better and fully buildable today. AI that drafts the report before you ask, then waits. AI that watches a process and speaks once, when the one important anomaly appears. Intelligence used to reduce the number of moments technology demands from a human — that's the product frontier I find worth working on.

The test I apply to everything I build now, AI or not: after a month of use, does the user think about this tool more or less than they did in week one? Less means it's working. More means I built another small alarm clock, and the world has enough of those.