Steve Jobs Killed My App

What twenty years on Flash taught me about building on someone else's frontier.

There are plenty of reasons to be wary of building a business on today's AI tools. Here is the one I am most afraid of, because it has already happened to me once.

In 2001 I shipped a residential real estate marketing SaaS application called Inkswitch MarketSpace. Its entire front end was built in Flash: layered, modular visual elements with behavior attached to them, stateful panels and buttons, a responsive interface that felt closer to a desktop application than to the static web pages of the day. It was a full AJAX-style experience years before anyone was using the term. The product ran for nearly twenty years, and as of 2019 it served more than 30,000 clients across a dozen companies, from Virginia to Hawaii.

Then, in 2020, it simply stopped working. Adobe ended support for the Flash Player plugin, browsers blocked it, and the interface that had carried my business for two decades vanished.

What Actually Happened

I have sometimes said, half-jokingly, that Steve Jobs killed my business by refusing to support Flash on the iPhone and iPad. I maintain that this is essentially true, though it is incomplete and perhaps a little unfair. Flash had real problems, and they were part of the reason Jobs rejected it: security holes, performance costs, battery drain, and an architecture designed for a mouse rather than a fingertip. It was also proprietary and not owned by Apple, which was never going to work for the most important new class of products Apple had ever shipped.

The plugin would likely have died of natural causes eventually. But the decisive acceleration came from a deliberate platform decision by the company that controlled the defining computing platform of the era, and the rest of the industry fell in line far faster than it otherwise would have. Adobe abandoned mobile Flash and, years later, announced the end of life. I spent a year and over $120,000 between 2017 and 2018 trying to replace the WYSIWYG PDF generator and a dozen other heavily visual tools I'd built in Flash with the current state-of-the-art HTML and JavaScript, but it was too little, too late.

The point is not that Jobs was a villain or that Flash deserved immortality (my editor insists I say this). The point is that I had built the distinctive experience of my product on a layer I did not own and could not replace, and when its owner walked away, my product died with it.

Skip Intro

Apple's decision was the killing blow, but the patient was already unpopular, and that matters more than people remember.

By the mid-2000s "Skip Intro" was a punchline. Every agency website opened with a thirty-second animated splash nobody asked for. Flash ads blinked and shouted and followed you around the page. Flash video spun up your laptop fan and ate your battery. Browsers added click-to-play and users installed blockers specifically to keep Flash from running at all. None of this was the fault of the technology as such; my application was not a splash page, and my customers loved it. But by the time Jobs published his open letter in 2010, the public had already decided what Flash was: bloat, intrusion, something to be skipped. He was pushing on an open door. A decision that would have caused an uproar in 2003 got a shrug in 2010.

I see the same reflex forming around AI right now, and much faster. "Made by humans" is becoming a badge people put on their work. Companies that announced AI-first strategies have walked them back after customer revolts. Illustrators, musicians, and writers are organizing against it. "Slop" has entered the language as the word for what these tools produce by default. Whatever you think of the merits, a meaningful share of the public is learning to skip.

No vendor can protect you from that. Anthropic, OpenAI, and Google can promise API stability until the end of time; they cannot promise that your customers will still want what the API makes. And when a platform owner or a regulator eventually moves against an unpopular technology, public distaste is what makes the move cheap.

Rented Capability

A large share of today's "AI applications" are thin interfaces wrapped around frontier models delivered through APIs. Their core capability lives behind someone else's service, often at promotional prices that will not last. A provider can raise prices, deprecate the model you tuned against, or change a safety policy in a way that quietly breaks your use case. A government can restrict access by executive action. It does not take all of these; any one can strand a product whose uniqueness sits on top of the external capability.

An API is more contractual than a browser plugin, and switching providers or self-hosting open weights is far easier than rewriting a Flash interface. The fragility is the same, though. Modern software is built from components we do not control: cloud, databases, auth, payments, and now language models. The question is degree and centrality. When the external layer is a commodity utility, the risk is manageable. When it is the source of the product's differentiated value, the risk is existential. Flash was not a utility for my application; it was the presentation layer. Many AI features sit in the same position today. They are the reason the product exists, not an enhancement to it.

I did not set out in 2001 to create a single point of failure. Flash was simply the best way to deliver what my customers needed, and for years the bet paid off. The risk only became obvious when the support disappeared. That is how these dependencies work. They look like infrastructure until they don't.

So treat the dependency as a live risk rather than a settled foundation. Watch the external layer's technical trajectory, its economic terms, its leadership, and the public mood around it. Keep a credible alternative and a realistic pivot plan. Abstraction layers, multi-provider architecture, ownership of the workflow and user experience, and periodic "what if this disappears tomorrow" drills are not overhead. They are the rent.

They Call Them "Frontier" for a Reason

A frontier is unsettled ground. The map gets redrawn, claims get contested, and nobody's title is guaranteed. The companies that survive the next transition will be the ones that kept the critical parts of their product under their own control, or at least built so that no single external decision could erase the whole thing.

My Flash application lasted almost twenty years and created real value for real customers. Its end was not a personal failure so much as a structural one, and the same structure is now visible in a new generation of products. The names will keep changing. The exposure will not. Build accordingly.