Part 1 of 3 · GMTech
The dominant story right now is that AI is going to kill software, but that story is wrong in a specific and important way. What AI is killing, is the cost of software. That’s a very different thing and the implication is bigger.
The democratization gap Andrew Ng saw coming
Andrew Ng has been making this case for years (TED):
The pizza store down the street generates valuable data, has obvious problems AI could solve, and has been priced out of every solution. Hiring an AI team for one shop has never made sense economically.
He called this the long-tail problem: millions of small businesses, each with unique needs, none individually large enough to justify custom software.
His parallel was literacy. A few centuries ago, reading was the priesthood’s job. Everyone else relied on a few people to interpret the world. AI today is in the same place — controlled by a small number of tech giants, with everyone else dependent.
There’s a hard ceiling on how democratized the model layer can actually be in 2026. Training a frontier model costs millions. Running inference at scale is now part of the power grid conversation. Owning a model takes capital and compute almost nobody outside a handful of labs can clear.
That kind of democratization is probably a fusion reactor away.
But the pizza store doesn’t need to own a model. It needs better demand forecasting, smarter inventory, less food waste. And in 2026, those outcomes don’t require training anything from scratch. They require software that could have been built a decade ago, but never was because the margins never justified it.
That’s what’s getting democratized right now.
Not AI, but the software that it enables.
What this looks like on Main Street
Jasmine Sun’s WSJ piece is a useful floor marker for how much easier this has gotten. A non-coder can build a personal dashboard over a weekend. Software built uniquely for one person. This is incredible. I love to see new people exposed to the thrill of building their ideas into reality.
… but the floor isn’t the ceiling.
A personal dashboard built over a weekend is one thing. Software fitted to a real business: its actual workflows, its specific data, twenty years of operational quirks living in spreadsheets. This is a different problem, with a solution that provides transformational value for that business.
Clive Thompson’s NYT story & Daily interview had another great example: A concrete company, $50 million a year in revenue, running its operation on three Excel spreadsheets and a Windows XP machine the team is afraid to update. The kind of business where custom software is obviously useful and where the attached price tag has always made it inaccessible. Like millions of businesses.
That math just changed. Software stops being something precious and rare. Every business is now a candidate for high quality software that fits their unique needs.
Why this is the paradigm shift
When commercial computing first took off in the 1950s companies couldn’t afford to, and didn’t have the expertise to, build their own software. So a generation of services companies, like my former employer Computer Sciences Corporation, emerged to build and maintain the systems.
That model held for over half a century. The price of custom software was high enough that buying off-the-shelf SaaS was the only realistic option for anyone smaller than the Fortune 1000.
AI is collapsing that price. Not by 20%. By orders of magnitude. Which means a software services company in 2026 isn’t the same shape as the consultancy of 2005. Higher margins. Lower price points. Smaller teams, more projects. Software that fits the business instead of forcing the business to fit the software.
This is the future of product-enabled software services.
What this doesn’t solve
Tools alone don’t close this gap.
The distance between a business that’s curious about AI and one that’s actually built to use it is gigantic and expanding faster every week. There is a platform modernization project of different scale for every company in the world right now. More thoughts later. Stay tuned.
— Bennett Bakke, Founder, GMTech