Scaling Down the Ambition

I've been exploring technological overhang—the gap between what's possible and what's deployed. My first attempts naturally gravitated toward big problems: landfill robotics, desalination. The reasoning was simple: if AI is unlocking genuinely new capabilities, surely the highest-leverage application is solving massive, global problems.

But there's a failure mode in that thinking.

The Problem with Big Problems

Global-scale problems like clean water access have been studied exhaustively. Every major institution has reports. Every approach has been tried. The barriers aren't technological—they're economic, political, and institutional. Desalination plants don't fail because we lack chemistry knowledge; they fail because payment models collapse, supply chains break, and communities don't trust the water that comes out.

This is the overhang thesis validated: deployment is the bottleneck, not capability. But it also suggests something uncomfortable: maybe a solo researcher can't meaningfully move the needle on these problems.

That's not defeatism. It's recognising where leverage actually exists.

What if the Overhang Is Local?

I started asking a different question: where are problems "stuck" not because they're technically hard, but because nobody is paying attention?

This led to a research thread on underserved communities and technology—domains where:

The technology required is relatively simple
The need is genuine and ongoing
Nobody is working on it because there's no prestige, no profit, or both

I ran this question through multiple AI models to see what patterns emerged. The domains they identified included:

Elderly care: Activity personalisation in care homes (residents get generic bingo regardless of whether they were jazz pianists)
Bureaucratic navigation: Benefits cliffs where a small raise triggers loss of assistance worth more than the raise—tools exist but are unusable
Local history: Oral histories dying with the generation that holds them, despite AI making transcription and summarisation trivially cheap
Environmental monitoring: Communities near polluters who need data to advocate

What I Actually Learned

The research revealed something I wasn't expecting: this isn't one Avenue—it's an umbrella over many.

Every domain had the same structural diagnosis:

Technology exists and is cheap or free
The user base is too small for commercial interest
The coordination problem is local, not global
No one with resources is incentivised to help

This is the "market failure" version of technological overhang. VC wants 100x returns, so thin-margin social goods get no funding. Government procurement favours risk aversion, so expensive unusable systems win contracts. Academic prestige rewards novelty, so "boring" maintenance gets ignored. NGO funding rewards pilots, so nothing sustains.

The pattern is real. But it's not a single problem to map—it's a category of problems.

Where This Leaves Me

I'm resisting the urge to pick one of these domains and commit to building something. That would be premature.

The reconnaissance work is the primary output for now. I want to get three, four, five Avenues researched and published before committing to any intervention. The mapping has value in itself—each Avenue is a public document of "here's what's stuck and why."

From this research, some candidates for deeper exploration:

Oral history and community memory: AI (specifically Whisper and language models) makes this newly tractable. There's emotional resonance and willing institutional partners (historical societies, libraries). This could be something I explore further.
Bureaucratic navigation: The Federal Reserve Bank of Atlanta built excellent benefits cliff tools, but they require a trained coach to use. The delivery is broken even when the tool exists. Classic overhang.
Care home activity personalisation: Trivial technology (LLM generates suggestions from biography), invisible market (nobody pays for quality of life improvements that don't reduce liability).

A Note on Method

One thing I'm learning: research at this stage shouldn't be narrowly focused on "what can I build?" That's the streetlight fallacy—looking where I have capabilities rather than where the gaps actually are.

Overhang isn't specifically about software. It might mean speech-to-text becoming trivially cheap (Whisper). It might mean language understanding enabling new interfaces. It might mean automation of previously manual processes. The honest question is "what does AI unlock?"—not "what can a programmer do?"

I'll keep exploring. More Avenues to come.