Flash Was Installed on 99% of Computers on Earth. Then It Was Gone.

For about a decade, Adobe Flash was not a tool used to build the web. It was the web, in every respect that distinguished the web from a document.

The navigation menus. The video players — YouTube launched on Flash in 2005, and so did Hulu, Vimeo, and Dailymotion. The games, from Newgrounds through Farmville. The interactive ads, the virtual tours, the corporate training modules, the portfolio sites. If you used the internet between 2000 and 2010, you used Flash constantly, whether or not you knew the name.

By 2009 it was installed on an estimated 99 percent of internet-connected desktop computers.

Ninety-nine percent is not dominance. It is totality. And on December 31, 2020, Adobe ended support, browsers blocked it outright, and an entire era of interactive content became unreachable in a single night.

Nothing about Flash stopped working. The ground it stood on moved.

The Layer Underneath

Flash's strength was that it bypassed the browser. Rather than depending on inconsistent, slowly evolving browser capabilities, it ran inside its own plugin — a controlled runtime that behaved identically everywhere. In an era when browsers agreed on almost nothing, that was an enormous advantage, and it is why Flash won.

It also meant Flash existed at the pleasure of whoever controlled the platform it plugged into.

In 2010 Apple declined to support Flash on iOS, and Steve Jobs published an open letter laying out the case: battery consumption, security, performance, touch interfaces designed around hover states that phones do not have, and — the real argument — that Apple would not let a third party own the layer developers built on top of.

Every criticism in that letter was technically defensible. None of them were why it mattered. What mattered was that mobile was becoming the majority of web traffic, and the largest mobile platform simply declined to carry Flash. Overnight, "works everywhere" stopped being true, and "works everywhere" was the entire proposition.

Meanwhile HTML5, CSS3 and JavaScript absorbed the capabilities Flash had existed to provide. Video, animation, canvas drawing, audio — all of it moved into the browser itself, as open standards nobody had to license or install.

The killer was not that HTML5 was better. For years it was worse: harder to author, less capable, inconsistent across browsers. It won because it was native to the platform, and Flash was a guest.

That is the general mechanism. Technologies rarely die because something better arrives. They die when the substrate they depend on shifts, and their central advantage becomes the thing that makes them impossible.

Network Effects Run Backwards

The second death pattern is faster and less merciful, and social platforms demonstrate it best.

MySpace. Friendster. Google+. Vine. Each was, at its peak, treated as infrastructure — not a product people liked but a fact about how communication now worked. Each emptied out at a speed that still surprises people who watched it happen.

The reason is that network effects are symmetrical, and almost nobody plans for the second half.

On the way up, each new user makes the platform more valuable to every existing user, so growth compounds and competitors cannot get traction. That is the moat, and it is real.

On the way down it runs identically in reverse. Each departure makes the platform slightly less valuable to everyone remaining, which makes the next departure slightly more likely. The people who leave first are the most engaged creators — the ones whose content gave everyone else a reason to open the app — so the decline in value is far steeper than the decline in headcount.

And the metrics lie throughout. Registered accounts stay flat, because nobody deletes anything. Monthly actives decline gently. What has actually collapsed is the thing nobody has a dashboard for: whether people believe this is where the conversation is happening. By the time a leadership team is looking at genuinely alarming numbers, the belief has been gone for a year.

BlackBerry shows the same dynamic from a different angle. Its defensibility was enterprise IT: secure email, physical keyboards, device management, and a procurement relationship with every large organisation. Then employees started bringing iPhones to work and demanding support for them, and the moat turned out to be a decision made by people who had stopped being the customer.

The Slow Version

Not every technology death is dramatic. The most expensive ones are the opposite.

Enormous quantities of critical infrastructure — banking core systems, insurance policy administration, government benefits, airline reservations — run on platforms designed decades ago, in languages with a shrinking pool of practitioners. These systems are not failing. They work, often with reliability that modern replacements struggle to match.

They are dying by attrition. Every year, the number of people who understand them decreases, and there is no pipeline. The knowledge is not in documentation; it is in the heads of people who have been maintaining the thing for thirty years and are retiring.

The failure mode here is not obsolescence. It is a system that still works and can no longer be changed — where the cost of a modification rises each year until, effectively, none can be made. Rewrites are attempted, run long, and get cancelled, because the specification exists only as the behaviour of the running system.

What Actually Signals Death

Market share is the worst possible indicator, because it is highest immediately before the collapse. Flash was at 99 percent two years from irrelevance. BlackBerry's revenue peaked in the middle of its decline.

Better signals, roughly in order of usefulness:

New developers stop choosing it. Not existing users leaving — new projects not starting on it. This precedes every other visible signal by years, because it determines what exists in five years.

The platform layer beneath it starts to shift. A new dominant device class, operating system, or distribution channel that treats your technology as optional rather than assumed.

Talent becomes cheap and then unavailable. Wages fall as demand drops, then rise sharply as supply disappears entirely. The second phase is the emergency.

The best argument for it becomes switching cost. When the honest case is "replacing it would be expensive" rather than "it is the right tool," the technical argument has already been conceded.

Its core advantage becomes a liability. Flash's plugin isolation. BlackBerry's enterprise control. This is the reliable one, and it is what makes the pattern so hard to see from inside: the thing you are best at is the thing that kills you, and it is still working right up to the end.

The Technology Collapse Pattern: How Dominant Technologies Die runs the cases — mainframe to PC, BlackBerry and Palm, Kodak, Flash, the social media graveyard, dot-com infrastructure, crypto, and the legacy enterprise systems still running — plus the eight signs and a transition playbook.