The Complexity Cartel

Enterprise software & Enterprise AI has a dirty secret: the solutions have become harder than the problems.

I call it the Complexity Cartel. Not a conspiracy. A gravitational pull. The quiet, compounding preference for intricate architectures that promise optimization and deliver entanglement. It’s the default setting of our industry and almost nobody names it.

ERP and transactional systems were designed to be the backbone of modern business. Keys to efficiency. What many became instead: permanent technical projects that can’t keep pace with the operations they were built to serve. Layer on layer. Integration stitched to integration. Each one justified in isolation. None questioned as a whole.

Now Enterprise AI is running the same playbook in newer language. Narrow semantic models scoped so tightly each one solves a sliver of a problem. Fleets of agents that need their own orchestration layer just to stay in sync with each other. Massive data engineering projects with multi-year roadmaps before a single business user sees a single insight. RAG pipelines. Vector databases. Fine-tuning loops. Evaluation frameworks to evaluate the evaluation frameworks.

Before anyone pauses to ask “what business question are we actually answering?”, the stack is six layers deep and the original problem is buried under its own infrastructure.

The pattern repeats every cycle. A complex solution arrives looking thorough, promising, inevitable. It addresses the original problem, then quietly outgrows it. Soon you’re not solving the problem anymore. You’re maintaining the solution. ERP did it with modules. AI is doing it with models, agents, and pipelines. The technology rotates. The trap doesn’t.

This is a first-principles failure. When the system is more intricate than what it was designed to address, something has inverted. The tool is no longer serving the work. The work is serving the tool.

The instinct in enterprise is always to add. Another model, another agent, another pipeline, another phase on the roadmap. But the highest-leverage move is almost always subtraction. Strip it back to what the business actually needs answered, and solve from there.

Simplicity is not a shortcut. It’s a discipline. Build solutions that scale to real needs, not imagined ones. Solutions that move at the speed of the business, not the speed of the implementation roadmap.

Here’s what frustrates me most. The real promise of AI was never to add another layer to the stack. It was to collapse the stack entirely. To absorb the complexity so the business never has to touch it. To make the machinery disappear and leave only the answer.

That promise is still available: This is why we built Maxa.

But only if we stop using AI to build more complexity and start using it to eliminate it.

The Complexity Cartel survives because no one stops to ask one question: is this solution still simpler than the problem?

Start asking.