I’ve been thinking about software differently lately.
For decades, software has been built on a simple promise: simplification. Take something complex—document editing, website building, event management—and make it accessible to everyone. Microsoft Word simplifies writing. WordPress simplifies web publishing. Signup Genius simplifies event coordination.
The business model is clear: build once, sell to millions. Users accept the limitations because building custom software is expensive and time-consuming.
But AI breaks this equation. And I think we’re only beginning to understand what that means.
The Hidden Cost of “Simple” Software
Let me share a concrete example.
Every robotics competition season, I create floor plans for our events. I use Microsoft Visio—professional software designed for exactly this kind of work. Feature-rich. Industry standard. Supposedly time-saving.
Except every single time, I spend half a day entering team numbers, adjusting layouts, fighting with the interface. The software does many things. I need it to do one thing. But to get that one thing, I wade through everything else.
Or consider WordPress. I’ve spent countless hours on formatting, plugins, updates, security patches. The platform can do almost anything. But the maintenance cost of “almost anything” is enormous—even when you only need a small fraction of its capabilities.
This is the dirty secret of traditional software: the simplification it offers often creates new complexity.
You trade one kind of work (building from scratch) for another kind of work (learning the tool, working around its limitations, maintaining it over time).
The 80-20 Problem
Here’s a pattern I’ve noticed: 80% of users only use 20% of a software’s features.
I’m one of those 80%. The software I use has dozens of features I’ve never touched. Menus I’ve never opened. Capabilities I’ll never need.
But I absolutely need that 20%. Without it, I can’t do my work.
So I’m stuck. Pay for 100% of the software. Use 20%. Spend significant time navigating around the 80% I don’t need.
This felt like an unavoidable tradeoff—until AI changed the math.
What If Software Was Disposable?
Here’s the shift I’m experiencing:
Instead of using a general-purpose tool that does many things adequately, I’m increasingly using AI to build single-purpose tools that do one thing perfectly.
Take our recent judging appointment system. Previously, we used Signup Genius or similar platforms. They work. They’re “simple.” But every event requires manual setup. The interface never quite fits our workflow. We compromise constantly.
This year, I asked AI to build a custom booking system. Exactly our requirements. Nothing more, nothing less.
Initial build time: about a day of iteration.
Time to set up a new event now: 5 minutes.
The software does one thing. It does it exactly the way I need. And if requirements change next season, I can generate a new version rather than fighting with configuration options.
The software is disposable. The capability is permanent.
The Real Cost Comparison
Traditional software:
- License/subscription: ongoing cost
- Learning curve: hours to days
- Maintenance: continuous
- Limitations: constant workarounds
- Updates: hope they don’t break your workflow
AI-generated custom software:
- Initial build: hours to a day
- Learning curve: minimal (you designed it)
- Maintenance: regenerate when needed
- Limitations: only what you accept
- Updates: rebuild to your new specs
For complex, enterprise-scale software, the traditional model still makes sense. You can’t AI-generate SAP or Salesforce.
But for the thousands of small tools we use daily? The calculation has flipped.
What This Means for the Software Industry
If individuals can generate their own software, what happens to the SaaS companies selling simplified tools to millions?
I think we’re heading toward a bifurcation:
Category 1: Platform software Large, complex systems that manage critical data, integrate with many services, and require enterprise-level reliability. These get more valuable, not less. AI makes them more powerful.
Category 2: Task software Small, focused tools that help people accomplish specific tasks. These are increasingly replaceable by AI-generated alternatives customized to individual needs.
The middle ground—software that’s too simple to be a platform but too rigid to compete with custom generation—is going to struggle.
The New Software Engineering
This isn’t the end of software engineering. It’s a transformation.
Software engineers will still build the platforms, the infrastructure, the complex systems that AI-generated tools run on top of.
But they’ll also need to think differently about what “software” means:
From products to capabilities. Instead of shipping a finished product, ship the ability to generate products. AI models, templates, building blocks.
From features to flexibility. Instead of adding more features, make it easier for users to get exactly what they need—even if that means generating something custom.
From maintenance to regeneration. Instead of patching and updating, make it trivial to rebuild from scratch with new requirements.
What This Means for You
If you’re a knowledge worker drowning in software that doesn’t quite fit:
Start small. Pick one annoying workflow. Ask AI to build you something better. You might be surprised how quickly you get something useful.
Think disposable. Don’t try to build the perfect, permanent solution. Build something that works now. Rebuild when your needs change.
Value your time differently. An hour building a custom tool might save you hundreds of hours of workarounds over the next year.
Question your subscriptions. How many SaaS tools are you paying for that AI could replace with something more tailored to your needs?
The Bigger Picture
We’re witnessing a fundamental shift in what software is.
For fifty years, software was a product. Someone built it, you bought it, you adapted your workflow to its design.
In the AI era, software is becoming a capability. You describe what you need, AI generates it, the tool adapts to you.
This is genuinely disruptive—not in the overused startup sense, but in the original Christensen sense. It changes who can create software, what software costs, and how long software lasts.
The companies that understand this will build platforms and tools that embrace generation rather than fighting it.
The companies that don’t will find their “simplified” solutions losing to custom alternatives that are simpler still—because they do exactly what the user needs and nothing more.
Software used to be something you bought. Now it’s something you describe.
The implications are still unfolding. But I’m convinced of one thing: the era of accepting software that “kind of” works is ending.
In the AI age, “good enough” isn’t good enough anymore. Not when “exactly right” is a conversation away.
Have you started building your own AI-generated tools? I’d love to hear what’s working—and what isn’t.
