Why Narrow AI Tools Beat General AI Assistants

Most people building with AI right now are building the same thing.

A chat assistant. A summarizer. Something that drafts emails. Broad, generic, impressive in a demo, forgotten in a week.

I build the opposite. Narrow tools that do one specific job inside one specific workflow, and nothing else.

Here’s the thing nobody tells you: narrow is harder to build and easier to trust.

Take any role where you’re tracking dozens of moving relationships or accounts at once, more signal than any one person can process manually. The generic answer is “use an AI assistant to help you stay on top of things.” I tried that. It was useless. Too broad, too generic, answering questions I didn’t ask in a tone that didn’t match how I actually think through a decision.

So I built something narrower instead. A tool that does exactly one thing: read recent activity across every relationship I’m tracking and sort it into a small number of clear buckets — act now, follow up, re-engage, leave alone. That’s it. No chat interface. No “how can I help you today.” Just a sorted list, every morning, before I’ve had coffee.

It sounds almost too simple to be worth building. That’s exactly why it works.

A generic AI assistant has to be good at everything, which means it’s mediocre at the one thing you actually need. A narrow tool only has to be right about one decision, made the same way, every time. You can test it. You can trust it. You can tell exactly when it’s wrong, because there’s only one job for it to be wrong about.

The unglamorous part nobody mentions: most of the work isn’t the AI. It’s the plumbing. Which data source do you trust more when two systems disagree about what’s actually happening? How do you filter out noise — the automated notifications, the things that look like signal but aren’t? That’s not a prompt engineering problem. That’s a judgment call about how you actually work, encoded once, so you don’t have to make it fifty times a week.

That’s the real shift AI enables. Not “ask a chatbot for help.” Codify your own judgment into something narrow enough to actually rely on.

Most people are still building chatbots. I’d rather build the thing that makes one decision extremely well, every single day, without me having to think about it at all.