The world doesn’t need more AI. It needs better AI.

We use AI. We also get tired of it.

You ask a simple question and get ten pages of text. Somewhere in there is the answer, but first you have to read an introduction and a list of things to consider. AI went from saving us time to something we spend time digesting.

Not every idea you have is brilliant, but AI tells you they are. AI even writes your emails and notes until nothing sounds like you anymore, and it’s taking over work that used to teach us something or that we actually enjoyed doing.

And there’s more of it every week. More apps, more agents, more AI-written content, more tools to manage the other tools. It feels productive. A lot of the time it isn’t. One study found experienced developers were 19% slower using AI, but they believed they were 20% faster.

The question we’re trying to answer.

How can AI actually help us, without taking over the parts of life and work that make us human?

We think it comes down to using AI on purpose. Let it help where it really saves time. Keep your own judgment, your own voice and your own skills. And know when doing it yourself is the better choice.

What we’re researching.

When should AI do it, and when should you?

Some work is worth handing off. Some is worth doing yourself, since that’s how you learn or because you actually enjoy it. We’re trying to figure out where that line is.

Less managing, more doing.

Right now using AI often means writing prompts, checking output and fixing mistakes all day. We’re looking at ways to hand off work that don’t turn you into a full-time AI manager.

Busy isn’t the same as productive.

AI makes it easy to feel like you got a lot done. We want honest ways to check whether you actually did.

Keeping agents in check.

AI agents are getting access to our email, files and accounts, and acting for us. More on that below.

We’re handing AI the keys.

AI agents can now read your email, open your files, send messages, buy things and run code for you. We rely on them more every month. And right now the usual approach is to give them access to everything and hope for the best.

We don’t think that should be the default. We’re researching how to give agents only the access a task needs, how to keep them from acting on instructions hidden in what they read, how a person can see and undo what an agent did, and how to contain systems that may eventually be more capable than the people supervising them.

Part of the work is simply making people aware. Agents can be genuinely useful. People should understand the risks before handing over the keys, not after.

Why there’s no product here yet.

Building is the easy part now. We’ve built and thrown away several prototypes already. The hard part is deciding what deserves to exist, because adding one more AI tool to the pile isn’t a contribution by itself.

So every idea has to answer a few questions before it goes further:

• Does it leave people more capable, or just more dependent?
• Does it save real time, or only feel like it does?
• Does it keep people’s data and decisions in their hands?
• Would anyone miss it if it went away?

Most ideas don’t make it. That’s the point of testing them.

We can’t figure this out by ourselves.

The best ideas so far have come from conversations, not from sitting at a computer. If any of this hits home, or you think we’ve got it wrong, reach out. We’d like to talk.

Reach out