We're in 1905: Why Electricity (Not Dot-Com) Is the Right AI Analogy - Freestyle Friday, 4/17/2026
TL;DR
AI looks more like electricity than dot-com — Joe Reis argues the better analogy is electricity’s 40-year rollout from the 1880s to the 1920s, where real productivity gains only showed up after companies redesigned operations instead of just bolting the new tech onto old systems.
Buying AI tools does not make a company “AI native” — He compares today’s “we got ChatGPT/Copilot subscriptions” mindset to the old “just put up a website and you’re an internet company” mistake: useful, maybe, but nowhere near embedding AI into the operating system of the business.
The moat problem is already hitting AI startups — Reis says wrapper companies are losing funding because there’s “really no defensible moat,” and he expects hyperscalers and foundation-model companies like OpenAI and Anthropic to look more like dot-com-era ISPs, with margin collapse and utility economics over time.
Most AI failures are organizational, not technical — Citing an MIT study saying 95% of AI projects fail, he stresses the issue is culture, workflow redesign, adoption, and training—not model capability—and says the real bottleneck is companies, not the tech.
Mandated AI usage is theater, not transformation — His example of managers forcing employees to use Copilot or get fired is, in his view, the modern equivalent of bolting electricity onto a steam engine: compliance may go up, but the business still won’t capture the technology’s full value.
The winning pattern is to automate toil, not threaten people — He tells the story of a founder running a 20- to 30-year-old business who is successfully using Claude Code to rewire systems because the company chose to change how it works and frame AI as a way to eliminate hated tasks, not eliminate employees.
The Breakdown
Tokyo walk-and-talk, and the big question on his mind
Reis opens from a crisp morning in Tokyo, fresh off the heat in Thailand, and uses the walk to frame the week’s obsession: where exactly are we in the AI rollout? He agrees with Malcolm Hawker that “the ground is shifting very quickly under our feet,” but immediately adds the harder truth — even when technology moves fast, organizations usually don’t.
Why the dot-com comparison only goes so far
He acknowledges why everyone reaches for the dot-com analogy: crazy valuations, easy fundraising, and a lot of “pandemonium and silliness.” But he says today’s AI startup landscape is already exposing the weakness of thin wrappers around APIs, with money concentrating into a smaller handful of companies as defensibility disappears.
OpenAI, Anthropic, and the ISP analogy
One of his sharper comparisons is that the big foundation-model players and hyperscalers resemble dot-com-era internet service providers more than unstoppable monopolies. In his telling, they may do the heavy lifting, but over time they risk margin collapse and utility-like economics, while consumers capture most of the benefit.
A ChatGPT subscription doesn’t make you an AI company
Reis says we’re repeating the old mistake of thinking surface adoption equals transformation: back then it was “we have a website,” now it’s “we have Copilot.” He hears constant complaints from employees stuck with weak, sandboxed corporate AI tools that aren’t integrated into real workflows, so the technology may be present on paper but absent where work actually happens.
The electricity analogy is the real one
This is the core pivot of the video: AI today, he says, looks much more like electricity in the industrial era than the internet in the late 1990s. Early factories tried to bolt electricity onto steam-era layouts and got little value; only after redesigning operations did labor productivity rise, and even then the payoff took decades — roughly from electricity’s introduction around 1880 to major gains in the 1920s.
Why adoption theater misses the point
He laughs at the story of executives buying Copilot for everyone and then threatening to fire people who don’t use it, because that’s usage theater, not transformation. His metaphor is memorable: forcing AI into a business without redesign is like demanding people use electricity by attaching it to a steam engine — technically new, strategically pointless.
The MIT study, tacit knowledge, and the real bottleneck
Reis points to an MIT study often summarized as “95% of AI projects fail,” and says the footnotes matter: the problem isn’t the technology, it’s organizational design. Companies say they want AI, but often won’t invest in training people to become AI-capable, and at the same time fantasize about replacing workers — ignoring the tacit knowledge locked in employees’ heads and setting themselves up for a painful “FAFO moment.”
A concrete success story — and his closing thesis
He ends with a more hopeful example: a founder friend running a 20-plus-year-old business is using Claude Code to rewire core systems successfully, precisely because the company was willing to change how it operates and use AI to remove toil rather than threaten jobs. That, for Reis, is the actual opportunity over the next several years and likely decades: not sprinkling AI on top, but rewiring the company’s operating system around it.