Jim Frederick

Every app you've used was a compromise between what you needed and what a few million other people needed. We called it good software when the compromise was tolerable.

The home screen on my phone will soon be almost entirely custom icons: apps I had an AI build for me, because each one is at least a little better than the one it replaced. That screen is also saving me at least $50 a month in subscriptions I no longer pay.

Prepackaged software is fine for the things I don't care much about. But for the things I do care about, messaging people, my health, finding information, my money, taking notes, and a dozen others, almost all of the apps stink. I don't want them anymore, and I don't want their fees.

Most people will still want an app that has already been decided for them, for at least the next five years, until developers get good at the feedback loop: using in-app analytics and events to reshape the app in real time for the person holding it. The evil cross-site cookie followed you around the web with baseball caps and skincare. A broader intelligence about your preferences could do the useful version of that, telling each app what your default experience should be and how it should change. When I buy a new phone in ten years and turn it on, there had better not be a single goddamn icon on that screen.

A factory wins by repetition. It stamps out identical parts and the whole point is that mine matches yours. Software made with AI goes the other way: each thing is shaped for one person, one week, one problem. Repetition stopped being the skill. Judgment is the skill now.

The first time software was free, we were teenagers with GeoCities pages and a view-source habit. Then it became an industry and the industry needed everything to be a product, and products need users, and users need onboarding. The feeling I have building with these tools is the one I had then. I can make the thing in an evening, and nobody has to agree that it's a business.

Picture no app store. You say what you need, out loud or in a few words, and something appears in the shape the moment calls for: a list, a map, a voice that answers back, a button on your watch. It lasts as long as the need does. The hard engineering is in the layer that chooses the shape and keeps your data yours. I've started to sketch what that layer looks like.

A plant has a life of its own. It grows, it adapts, it mostly sustains itself. What it asks of you is small and specific: water, light, a bit of nitrogen now and then. Software is starting to look like that. The agents write it and rewrite it. What it needs from a person is a thin, steady stream of judgment: are the security patches landing on time, do people still want this, is it growing in a direction worth keeping. Less a factory floor than a windowsill.

For thirty years the answer was at the point of sale, for the software itself. That's ending for anything one person can describe in a paragraph. Payment moves to the moments a homemade version can't cover: when the data has to come from somewhere, when the thing must not go down, when nobody on hand knows what good looks like, and when it breaks at two in the morning and someone has to answer for it. Businesses that sold the first thing will have to find out which of those moments they actually own.

Niklas Gruhn's Don't be a meat proxy told us to stop pasting the model's answer into Slack. Fair. Now look at a whole organization. Every white-collar person in it uses their own expertise to steer a model toward research, ideas, discoveries, and recommended decisions, then hands the result to the next person, who does the same. Most of what moves between us is an AI artifact with a human stamp on it. So what are we? Are we all holding the leashes of pets we only half understand, or is there a better metaphor? How do you build a healthy organization when nobody is sure how much of the work is theirs? And the harder question: how do you keep your self-respect inside it? I have guesses, not answers, and they get their own post.

For thirty years you could make a killing by rearranging zeros and ones. Move a number from one database to another, take a fee, call it a company. That trade is closing. When anyone can conjure the software, the software stops being the valuable part. What's left is the world. Make a physical thing that changes someone's day. Be the doctor who adds years to a life. Build something that has to hold weight. You can't be a ghost in the seams of reality and expect to be paid for it.

I don't mean to be such a harsh critic of software engineers. I lived in that world deeply and long enough to know that at least half the people in it happened upon it. Nothing wrong with that on its own. But they distracted the rest of us and made the work harder. Every extra voice in the room adds time and complexity. I wish everyone well, in their lives, their work, whatever they take up next. It's just hard to hear a profession whine about perfection and best practices and the ivory systems it upholds, and now, as the money thins out, the slow oncoming drone of complaint.