The human timeline

In the first post I wrote that it was both exciting and a bit nerve-wracking to watch my job turn into something I barely recognize. Three posts later, I have made up my mind about the nerve-wracking part.

The estimate

Ask a capable model today how long a task will take, and it still answers in human time. “This is a multi-week effort.” “Plan for two sprints.” Then the same model, running as an agent, does the work in fifteen minutes.

The model has not caught up with what it can do, and most of us have not either. That is the part that needs recalibrating: not how fast the agents go, but what we allow ourselves to imagine and build.

Pencils down

Last week at Rails World, DHH announced that 37signals has gone “pencils down”: writing code by hand is now the exception there. He described a quarter of a century of chiseling code by hand, and said we should look back on it “not with regret, but with joy of what was”.

I recognized myself. I loved the long, deep sessions, fighting a large codebase until the right solution came out and it finally worked. That pleasure is mostly gone. It may survive as a sport, something done for fun on a weekend, but it is no longer how software gets made, and that is actually good news.

The flow is still there. It is no longer one codebase held in my head for hours, it is making as much as I can with a dozen sessions moving at once: less serial, more parallel, a bit more frantic, but there is a flow state to find in that too. I would never put the genie back in the bottle.

What comes out of the factory

I am building five or six apps in parallel right now and trying to ship roughly one a week, mostly native apps for the Mac and iOS, because that is what I use every day and where selling software is the most straightforward. After more than a decade of shipping software by hand, I look at what comes out of the factory with the same eye I used to look at my own work. Some people may call it slop, but I do not think it is.

It is not perfect, for sure. These are early releases, with the rough edges every early release has, but that was true of every 1.0 I ever wrote by hand. It is still good software, and I know it has an order of magnitude fewer bugs than what I used to ship.

We taught rocks to think, and in a few years they went from autocomplete to building whole complex applications. For the first time in history we have a shot at building the right software for every need and every person out there. Not doing it would be the slop.

Scope, not speed

The more interesting consequence is not speed, it is scope. DHH put it well: “every dream, every annoyance, every paperclip, every frivolous feature suddenly within reach”.

For me, that is a 3D design tool in the territory of Fusion 360, started mostly for fun because KiCad never worked the way I wanted it to. A year ago, one person starting a CAD tool on the side would have been a joke. Today it is a reasonable weekend project that keeps going.

What makes it worth doing is not the code, it is the opinion. If you have strong ideas about how a piece of software should work, what you would actually want to use, that is now the scarce input. Other people probably want the same thing, and nothing stops you from building it anymore.

The next two years

The obvious question is whether any of this is worth anything. If anyone can ask Claude for an app, will anyone still pay for one?

I think both can be true. Custom software, generated on demand on everyone’s iPad, is coming, and sooner than most people expect. But there is a window before it arrives, a year, maybe two, where someone with taste and a working factory can build applications for people that they could not build for themselves. That is what I am spending mine on.

Makers of things

DHH described what comes next as “a new career, an exciting career, an energized career as a professional maker of things”. I think that is the right way to see it. The job is not disappearing, it is becoming a different one, and very small teams with very small budgets are about to make things that used to need a company.

Most people outside our industry do not realize yet how good these models have become, or what that means for any job built on thinking. We see it first because code is where the models are strongest. That is valuable information, and the right thing to do with it is to act on it: think bigger, start the things that were out of reach a few months ago, and switch over sooner rather than later.

The flip side is what happens to all the software that already exists.