No, it's imagination, same as it ever was.

Software to most of this discussion is a web app with a landing page, a pricing tier, and MRR. That's it. The frame is "product," the metric is "traction," and the canvas is "things people pay $9/month for."

But software is instructions that make matter and energy do things they wouldn't otherwise do. It's the most general-purpose tool humans have ever built. So let's actually think about what's underbuilt:

The whole damn physical world is barely instrumented. Agricultural systems, water infrastructure, building envelopes, soil health, local microclimates. There are farmers making irrigation decisions on vibes. Municipal water systems with no real-time leak detection. Buildings hemorrhaging energy because nobody's modeled their thermal behavior. These aren't apps. They're control systems, and they're mostly missing.

Fabrication and manufacturing are being transformed by CNC/3D printing but the software for designing things to be manufactured is still terrible (and inaccessible!). Generative design that accounts for material properties, toolpath constraints, assembly sequences. CAM software is where word processors were in 1985.

Scientific instruments. A spectrometer is mostly software now. So is a radio telescope. So is a seismograph. Every goddamn thing can be a thermometer (accidentaly!) The gap between "data sensor exists" and "useful scientific instrument" is almost entirely software, and most of that software is written by grad students in unmaintained Python.

Preservation. Some people are doing this with datamuseum.dk. But expand it: there are entire musical traditions, oral histories, craft techniques, ecological knowledge systems that exist in living memory and nowhere else. Software for capturing, encoding, and transmitting that knowledge barely exists. Not "an app for recording grandma," but formal knowledge representation of, say, how a master boatbuilder in Kerala selects wood by sound and flex.

Prosthetics and rehabilitation. This one is big for me personally! The gap between what a modern prosthetic limb could do with good software and what it actually does is enormous. Why are my eyeballs still chunks of plastic? Same for cognitive rehabilitation tools, speech therapy systems, physical therapy feedback loops.

Governance and collective decision-making. Every organization above 20 people is making decisions with tools that are basically "email plus meetings plus a shared doc." Formal deliberation systems, preference aggregation, transparent resource allocation. These are hard computer science problems that nobody's building because they don't have obvious MRR.

Tools for thought that aren't note-taking apps with backlinks. Actual reasoning aids. Argument mapping. Assumption tracking. Decision support that makes your thinking better rather than your typing faster.

The entire domain of formal verification applied to things that matter. Bridges, medical devices, voting systems, financial settlement. We have figured out how to prove some? software correct. We almost never do it for the software where correctness actually matters.

And that's me, one person, in five minutes. Every domain expert in the world is sitting on a pile of unsolved problems that software could address, and most of them have never talked to a programmer because programmers are busy building the next task management app.

Go talk to a nurse, a farmer, a building inspector, a food bank logistics coordinator. Ask them what's broken. I promise the answer isn't "nothing" and I promise nobody on ProductHunt is solving it.

PG wrote essays about this ffs! "Make something people want." "Live in the future and build what's missing." That advice didn't stop being true because AI made the building part cheaper. If anything it's more true now, because the building is almost free, which means the noticing is almost the entire game. You are skipping the noticing and going straight to the building, then wondering why nobody cares.

The number of hard things isn't going down. This thread can't see them because it's not looking at the world. It's looking at ProductHunt.

To make things concrete, in the last week I have been working on my open source speech synthesizer, rebuilding Klatt's ideas from the 1980 paper up to modern emotion/prosody work. Did you know the whole field went nuts for neural approaches in ~2018 and there's a whole shitload of interesting papers just sitting out there that nobody has ever implemented in a real system? Did you know that a bunch of people did research into what different human emotions sound like and now I can make a depressed speech synthesizer, or, scarily, one which sounds more honest to people?