The history of computing is a rhythm. Consolidation, then fragmentation, then consolidation again. Each cycle creates a new generation of companies, tools, and opportunities.

Mainframes concentrated everything in one room. One vendor, one machine. PCs shattered that — computing moved to every desk, and thousands of software companies filled gaps the mainframe vendors never imagined. Cloud consolidated again — centralized infrastructure, dominant SaaS platforms, the mainframe model with better UIs and a subscription fee.

AI is triggering the next fragmentation. And like every previous one, it's a leveling event between small and large businesses.

How Platforms Lose Their Soul

Every great software product starts with an opinion — a clear point of view about how work should flow. Users love it because someone who understands the problem designed the solution.

Then growth happens. More features for differentiation. More integrations to reduce churn. Product managers survey customers, figure out what most want, and build that next. The goal becomes checking boxes on RFPs and winning feature comparisons.

None of these are bad decisions individually. But they compound non-linearly. Every feature you add is a feature you stop curating. By version twelve, the opinion that made the product worth using has been diluted into something that offends nobody and delights nobody.

The product stops being opinionated. It starts being comprehensive. Those are not the same thing. What ships is the average of all customers' needs — the mean, not the fit. Every feature added for differentiation becomes a liability in complexity, maintenance, and diluted taste.

Why Fragmentation Is Back

Every consolidation cycle ended because someone made the alternative cheap enough. Mainframes dominated until microprocessors made PCs affordable. Cloud platforms dominate because building your own used to require a server room.

AI collapses the two costs that kept fragmentation impractical:

Building. Software that took a team months now takes days. A custom tool for one workflow is now justifiable because the build cost dropped by orders of magnitude.

Connecting. AI handles the integration work that made small tools impractical — moving data between systems, translating formats, maintaining connections. Not replacing the taste in each tool, but handling the mechanical work of making them cooperate.

Both barriers falling at once isn't incremental improvement. It's a paradigm shift.

The Jevons Paradox of Software

In 1865, Jevons observed that making coal more efficient didn't reduce consumption — it increased it. Software is about to follow the same pattern. As it becomes easier to build, we need dramatically more of it.

Think about every workflow never worth building software for. The fabrication shop on spreadsheets because no PM tool fits their scheduling. The consultant tracking finances in a notebook because every accounting app serves someone else's business. The team duct-taping three SaaS tools because none match the actual process.

All software opportunities now — not as products for millions, but as bespoke tools for specific people. The bottleneck shifts from building to knowing what to build. Domain expertise becomes the scarce resource.

Whose Loop Are You Compounding?

When you adopt a platform, every hour learning its interface compounds their flywheel, not yours. Switch vendors and that investment evaporates. The knowledge doesn't transfer. The data export is a CSV that maps to nothing.

Build a tool that fits your process, and every refinement compounds your operation. The workflow encodes your hard-won domain understanding. Generic tools produce generic results — a fab shop running the same software as a marketing agency manages projects the same way, which means neither does it well.

Bespoke tools reverse that. The tool gets better because you understand your problem better, not because a vendor shipped what 60% of surveyed customers requested.

Where This Goes

The old model: one vendor surveys customers, ships the average of everyone's needs. You fill out the feature request form and hope. The new model: builders assemble exactly what's needed from open-source projects, APIs, and focused tools. You're not a customer requesting features — you're a builder assembling capabilities.

We're going to need more software than we've ever had. For use cases that were never addressable. Built by people who understand the problem better than any platform vendor could. Connected by AI that manages the seams.

Every feature a platform adds is a liability. Every capability you choose to include in a purpose-built tool is an asset, because you selected it with intention.

This isn't a crisis for software. It's a golden age for builders.