Dark Software Factories are Cool. The Interesting Bit Is Everything Else
I recently glimpsed dark software factories in practice. Named after the lights-out manufacturing plants where robots build things in the dark. The lights are off because lights are for humans, and there are no humans.
I think about building software that way and what it means.
Not in a "AI replaces engineers" sense; that framing has enough people talking about it, and my bet is it's mostly wrong about what's actually shifting anyway. I mean it more structurally: what does a software organisation look like when the effort and time of building a feature trends toward trivial? When the bottleneck stops being the time it takes to build this and becomes something else entirely? Because that's the point of bottlenecks. We spend time removing them, only to find where the next one is.
So that's the thing I'm thinking about, because the software factory is coming. It's here. I'm trying to get my thinking ahead of that. About what comes after. About everything downstream.
The shift
The hardest ideas to execute are becoming trivial.
I don't mean "easier." I mean trivial, in the mathematical sense of the word, where something that used to require a team and a quarter now requires a prompt and a coffee. Things I would have estimated in weeks I now estimate in hours. Things I wouldn't have attempted at all because the ROI sat firmly in "juice is not worth the squeeze" are suddenly seductively viable.
The pace of new features will be hourly. It already is, in some places. The question is what that does to everything that was created and grew up around the assumption that features were expensive.
Three things that break
The human, both sides of the screen. A product manager used to be able to hold the product in their head. A designer could keep up with what was shipping. An engineer knew the codebase. None of that survives an hourly cadence. The mental model of "I understand this system" stops being achievable through attention and effort. It has to be achieved some other way, or abandoned. I don't yet know which.
And what about the end user? The end user is a human too, and they've been quietly relying on the same thing: a product that sits still long enough to learn. Muscle memory, the menu in the place it was last week, the workflow that became second nature. Hourly features break that contract. We're about to ask users to keep up with software that changes faster than they can form a habit around it. Nobody has solved that problem because nobody has had to.
The organisation. Companies are coordination structures that exist because building things together is hard and requires a lot of people pulling in the same direction. Standups, sprints, OKRs, roadmaps; these are scaffolding for a problem that's becoming smaller. What happens when the scaffolding outweighs the building? Most orgs will keep the scaffolding, because the scaffolding is what they know how to measure. The ones that don't will look weird and probably win.
The path from user to value. This is the one I find most interesting, and most underdiscussed. If features are hourly but the user with the need still has to find them, install them, learn them, trust them, this is where I think we've moved the bottleneck without solving the problem. The constraint shifts from supply to discovery, from build to match. The user with the need and the feature with the value are further apart than ever, even as both multiply.
What I want to figure out
The dark factory is the easy part. Or at least, it's the part that is being solved right now. The interesting part is what sits around it.
How does a human stay meaningfully in the loop when the loop runs faster than they do? Not as a rubber stamp, not as a bottleneck, but as something that adds something the factory can't.
How does an org stop being a coordination overhead and start being a taste filter, a values filter, a direction-setter? Because that's what's left when execution is free.
And how do we close the gap between someone who needs a thing and the thing that does it, when the thing exists, but the person doesn't know, can't find it, can't trust it, can't tell it apart from the ten thousand other things that also exist?
I have a hunch the answers don't look like the current answers. And I have a hunch about what some of those answers might look like. That the unit of software stops being the feature and starts being the need. That the interface between user and software gets a lot more active about meeting the user where they are, rather than waiting to be found. That the organisations that win won't be the ones pumping out features the fastest, but the ones that get fast and stay deliberate about value and outcome.
But that's the next post. For now: the factory is here. The interesting questions are everything around it.