Software finds the pattern. A person shows up.
A foundation CEO named the line between software and people. It is the same line I design to.
Notes on design, AI, and the tools that earn their place.
A foundation CEO named the line between software and people. It is the same line I design to.
I gave one of my AIs its own email address. Not so it can email people. So it stops using mine.
Most software built for relationships is built to talk. A donor experience lead asked for the opposite.
Most software is designed to make leaving hard. Reversibility is the feature.
Seth Godin drew a line between two ways organizations approach AI. One cuts. One builds.
Five words did more work than anything else we published this month. Trust is the list of things you commit to never doing.
Six months ago I did not think sovereign Canadian AI was realistic. Now it has changed the questions I ask.
In February I said dashboards were becoming scaffolding. Four months later, I stopped building one.
AI will not arrive in a nonprofit budget as one big line item. It will arrive as many small ones, and most will not replace anything.
A publisher added a Human / Agent toggle to its articles. Helpful, manipulative, biased or optimized? It could be all four.
The AI layer of the web is being built right now. What is still missing is the link.
We used to take pride in mockups that looked real. The bar has moved from pixels to content.
I asked an AI system to run a week of LinkedIn content. What it chose not to do taught me more than what it did.
Thirty minutes with Claude Design produced a full identity for a fictional foundation. Then it had to survive a week of real content.
The old rule was that you grow by hiring. That rule is breaking, and the same shift is coming for nonprofits.
Nobody is building the llm:// protocol yet. So I built a workaround.
I stripped away the features, the agents and the dashboard to find the single thing AI4Love does for people. It was not technical.
We solved one-click linking once for browsers. AI needs the same standard: llm://
I am a print and web guy. I wanted an animated product demo anyway, so I built it in Terminal.
I think faster than I type. A local voice-to-text tool that removes the slowest step.
A "zero-human company" of AI agents, pointed at real work. I became the manager of managers who managed nothing.
Not every tool ends up in the graveyard. This one doesn't try to be smart. It tries to be fast.
Thousands of simulated agents debating your problem. Hours of compute, minutes of insight.
I believe 2026 marks the end of the dashboard. We will stop designing places to look and start designing systems that respond.