· Systems · 10 min read

The SaaSpocalypse Is Upon Us. Again.

AI made writing software cheaper. It didn't make understanding reality optional. Seven Adobe alternatives, 56 RGB profiles, and the inconvenient difference between generating code and building a product.

AI made writing software cheaper. It didn't make understanding reality optional. Seven Adobe alternatives, 56 RGB profiles, and the inconvenient difference between generating code and building a product.

The Day Adobe Died (Again)

Recently, the internet buried Adobe again. At this point, the company should offer a funeral template with its subscription. This time, the occasion was ArtCraft: seven open-source applications targeting Adobe’s creative suite, built with AI by developer Brandon Thomas after frustration with Adobe’s cancellation fees. The viral version practically wrote itself: one man, one chatbot, one week. Decades of software development, apparently exposed as an elaborate employment scheme.

Microsoft Office was next. By the time I wrote this, WordCraft, GridCraft and DeckCraft already existed. Somewhere in Redmond, an executive presumably dropped his coffee. At this rate, Windows would be obsolete by Thursday.

There was, however, a small complication. Thomas himself described the initial releases as “super early alpha” and said reaching full parity would require time and help. The week-long framing isn’t an independently verified development log, either. Caveats rarely survive the journey from repository to LinkedIn. By the third repost, an experiment becomes a product, a product becomes a revolution, and somebody starts selling a course. Before cancelling every subscription, though, perhaps we should ask whether having an alternative is actually enough.

The Competitors That Forgot to Win

GIMP, LibreOffice, Shotcut, Blender and Linux have existed for years, providing serious, often free alternatives to commercial software. Apparently, nobody told Microsoft that LibreOffice had already destroyed its business model. These aren’t clones in the traditional sense. Their features and workflows differ from those of their commercial counterparts. But they are fully fledged applications capable of professional work, not weekend demos waiting for someone to implement the actual application.

The list extends across fields: Scribus for desktop publishing, Krita for digital painting, FreeCAD and OpenSCAD for different approaches to CAD. Pick a discipline and you’ll probably find an open-source project that’s been doing the supposedly impossible while everyone was busy announcing that it had finally become possible. Their existence demonstrates that a useful, mature alternative can exist without automatically displacing the incumbent.

People buy into workflows: colleagues who use the same tools, clients who send particular files, integrations, training, plugins and decades of accumulated organisational sediment, layered over habits picked up at school or university. Migration has costs even when the replacement’s price is zero. Sometimes the incumbent is better for the job. Sometimes everyone just knows where the bloody buttons are. AI can lower the cost of building a competitor, which matters enormously, but it doesn’t automatically lower every cost of becoming its customer. Before we even get to adoption, though, there’s another detail to establish: what, exactly, has been built?

Seven Applications, Three Buttons, and a Dream

I haven’t audited ArtCraft’s code, and I’m not going to declare seven repositories fraudulent because their development story annoyed me. A promising early application can contain substantial engineering and still be nowhere near a general replacement. What interests me is how much complexity a familiar interface can conceal.

Take file support. “Opens PDFs” sounds like a checkbox. PDF has specialised standards for archiving, professional printing and accessibility. JPEG includes multiple coding modes. PSD carries layers, masks, effects and metadata. Reading a file, displaying it correctly and saving it without quietly mangling someone else’s work are different achievements. Libraries can handle much of this, which is excellent, provided you establish what they support, integrate them correctly, and test the combinations your customers actually use. The complexity has acquired dependencies, not disappeared.

Colour offers a particularly entertaining example. While working on my book covers, I tried to get consistent colours from Amazon KDP and IngramSpark. My system had 56 colour profiles. Fifty-six. Just for RGB. There’s also CMYK, plus several other colour models I discovered only because I made the tactical error of reading the options in the ‘Convert Image Colour Space’ dialog. Apparently, even a dropdown can reveal how little you understand about reality.

All I wanted was for the same cover to look like the same cover, a demand the universe apparently considered impertinent. Screens emit light; paper reflects it. Profiles describe how colour values relate to colour spaces and particular devices. Even printing the same cover twice can become an engineering adventure. Your “Export” button is sitting on top of a small international diplomatic incident.

Now combine filters, masks, transparency, GPU drivers, operating systems, huge documents and undo histories. A filter working once says little about applying it after resizing a masked layer, undoing three operations, saving, reopening and continuing on another machine. Unlike a demo following a prepared route, users wander through the state space carrying scissors. You can reduce that space through good design, constrain it through scope, and explore it with automated tests. But eventually somebody must decide what correct behaviour means. A demo proves something can work. A mature product must survive the inventive ways people make it fail.

Now Do Accounting

Photoshop is merely a convenient example of a confidence that now extends to accounting, HR, logistics, ERP and document management. “Build me an accounting application. Make it modern. Add a dashboard. Use React.” Minutes later: forms, tables, a chart and a green notification saying “Success.” It looks reassuringly similar to software that knows what it’s doing.

Then you need to correct an invoice from a closed reporting period or handle a transaction whose tax treatment depends on jurisdiction. You need an intelligible audit history, appropriate retention rules and permissions that keep one customer’s data out of another customer’s search results. Somewhere in there, you also have to reconcile imports from the system everyone insists will be retired next quarter. Naturally, a narrow accounting tool needn’t support every jurisdiction or transaction. Defining those limits is part of designing it. Calling the result a replacement for an entire accounting platform quietly removes the limits while retaining the development estimate.

I wrote about this in Tesler’s Law: Complexity’s Enduring Burden. Some complexity belongs to the problem itself. We can move it, automate its handling, hide it behind an interface or choose a smaller problem. Tesler’s law is a design heuristic, not thermodynamics wearing a conference badge. Plenty of implementation complexity is accidental; AI can help eliminate it. It can also help analyse requirements and handle legitimate complexity.

Tax rules don’t stop applying because Claude generated the form, though. Someone still has to establish which rules matter and whether the result obeys them. Those boring parts aren’t the stuff around the product. They are the product.

Congratulations, You’re a Software Vendor Now

Buying SaaS also buys some bundle of ongoing work: hosting, updates, monitoring, backups, security and support. The quality varies. Some providers seem to interpret “managed service” as “we manage the invoices.” Generating your own application may remove the subscription, but the work still needs doing.

Who patches a vulnerable dependency? Who updates the business rules? Who restores the database, and how do you know the backup is usable? Who diagnoses an incorrect result that looked perfectly reasonable for three months? AI may help with every one of those jobs, but you’ll still need a way to notice they’re necessary and verify they’re done correctly.

Congratulations. You replaced a hypothetical $49/month subscription with a software maintenance department. The invoice just looks different. Sometimes that’s an excellent trade. Some subscriptions are overpriced; some internal tools deserve to be twenty lines of code. The calculation should include ownership, though. Having the source code makes you responsible for a new set of decisions. It doesn’t automatically make those decisions free.

The Part Where We Accidentally Insult Our Own Professions

Here’s the more interesting question: what are you claiming about your own field when you announce that its specialist software can be replaced in a weekend? Perhaps you’re missing much of its complexity: the exceptions, historical constraints, awkward dependencies and reasons why apparently redundant fields exist. The system looks simple because the difficult parts haven’t entered your model. Or perhaps the task really is that simple, which is fair enough, although declaring that your professional tooling amounts to three forms and a pie chart is an unusual way to advertise the sophistication of your profession.

There’s also a thoroughly sensible third possibility: you understand the domain and need only a narrow slice of it. Your invoice tracker doesn’t need multinational consolidation. Your image editor doesn’t need a print-production workflow. Your internal approval tool doesn’t need to become Salesforce, although give management six months. That is precisely where cheap implementation becomes valuable. You can finally afford a tool shaped around your actual needs.

The intellectual fuckup happens when “this handles my workflow” becomes “this replaces the category.” The missing requirements don’t announce themselves. An absent button is visible; an absent rule can remain invisible until it costs money. Solving your problem is an achievement. Assuming you’ve therefore solved everyone else’s problems is a diagnosis.

So, Is SaaS Actually Doomed?

Some vendors should absolutely be nervous. If your product consists of a database, a few forms and an enterprise subscription justified mainly by the former expense of writing those forms, cheaper implementation is a serious problem. Narrow workflows and internal applications are obvious candidates for rebuilding.

AI can also accelerate requirements analysis, testing, debugging and maintenance. Competitors and incumbents alike can use it, and a small team may accomplish things that previously required a large budget. None of my argument requires pretending otherwise. That can change build-versus-buy calculations, create price pressure and make entire products uneconomic. A challenger doesn’t need perfect parity to win customers whose needs it meets.

But establishing which customers, which needs and which guarantees remains essential. “Good enough for this market” is a product strategy. “All software is now worthless” is engagement bait with a valuation model attached. AI can commoditise implementation while domain expertise remains valuable. It can even help develop that expertise. You still need evidence that the important problems have been understood and handled.

The Code Was Never the Whole Job

I want more people to build software, experiment, replace an annoying subscription or make something wonderfully specific that no commercial vendor would bother selling. There’s enormous value in lowering the cost of trying. Seven early applications can be an impressive achievement and may become serious competitors. Neither possibility requires treating their existence as proof that decades of accumulated knowledge were unnecessary boilerplate.

The Unix tradition has long valued simplicity, and The Art of Unix Programming opens its chapter on complexity with a useful warning: “Everything should be made as simple as possible, but no simpler.” That principle travels remarkably well beyond tools for an operating system beloved by people who consider a blinking cursor a welcoming user interface. It applies to accounting, printing, logistics and pretty much anything else where reality has an opinion about the result. Making life easier and letting more people build things are excellent goals. The trouble starts when enthusiasm for the first half of the sentence makes us forget the last three words.

Simplify below what the task actually requires and you haven’t outsmarted the complexity. At best, you’ve borrowed against it. Repayment arrives as workarounds, years of special-case handling, downtime, lost data or regulatory fines. More likely, you’ll receive a thoughtfully diversified portfolio of all the above. You can deliberately support fewer cases and say so; that’s a smaller product doing its job. Promise the same job while quietly omitting what makes it work, and the apparent saving becomes somebody else’s future invoice. Reality offers credit. It does not offer debt forgiveness.

AI can write code faster than I ever could, generate an interface before I’ve finished my coffee, and implement things I wouldn’t know how to build myself. What faster generation doesn’t establish is whether we’ve asked for the right thing, covered the relevant cases or recognised the consequences of getting it wrong. And if your entire business plan depends on reality having become simpler because a chatbot has become better at producing code, well… good luck with that.

The clones and the claims

The ArtCraft launch, its Office follow-up, and the distinction between early releases and finished replacements.

4 sources

What file support hides

Standards and format features behind deceptively simple import and export buttons.

3 sources
  • PDF AssociationAccessed: 2026-10-19

    Maps the core PDF standard and specialised standards including PDF/A for preservation, PDF/X for print workflows and PDF/UA for accessibility.

  • JPEG committeeAccessed: 2026-10-19

    Describes the multipart JPEG 1 standard and its range of coding modes.

  • Documents preservation of layers, effects and masks in PSD, and differences between formats and bit depths.

Colour is not a dropdown

The device and colour-space relationships behind colour management.

1 sources
  • International Color ConsortiumAccessed: 2026-10-19

    Explains ICC profiles and colour reproduction across devices. The KDP/IngramSpark cover-production experience and 56 RGB profiles are the author’s personal anecdote.

Simple, but no simpler

Complexity, the Unix tradition and the limits of simplification.

2 sources
  • Adam Korga Adam KorgaAccessed: 2026-10-19

    Earlier essay on moving complexity between users, applications and platforms. Used as a conceptual cross-reference.

  • Eric S. Raymond Eric S. Raymond 2003Accessed: 2026-10-19

    Opens the chapter with “Everything should be made as simple as possible, but no simpler” and discusses the Unix tradition’s pursuit of simplicity. The article links the wording to this chapter without asserting that it originated in Unix.

Back to Blog

Related Posts

View All Posts »
Zawinski’s Law Was Too Modest

Zawinski’s Law Was Too Modest

Every program expands until it can read mail — and so does every company. A field guide to bloat, strategic diversification, and the vanishing brake that used to say "enough."

Every Optimization Creates a New Blind Spot

Every Optimization Creates a New Blind Spot

Every optimization creates a new blind spot. It's the unspoken tax of systemic design — and it's been killing projects, supply chains, and occasionally, entire populations.

Tesler's Law: Complexity's Enduring Burden

Tesler's Law: Complexity's Enduring Burden

Larry Tesler gave us cut, copy, and paste. He also gave us an uncomfortable truth: complexity never disappears. It just moves. The only question is who gets stuck with it.

AI: The New Steam Engine Trap

AI: The New Steam Engine Trap

Right now, the software industry is about to step on the same rake. Again. We’ve spent 150 years assuming that swapping one component for a faster one would magically fix the economy.