Vertical software in the AI era, where the moat is process engineering

Models are getting more and more powerful. That doesn't mean vertical software is dead. It means shallow software is dead. The real moat in the AI era lies in process engineering, and not in the model.

Editor's note. This article came out on February 20, 2026, a few weeks after a sharp drop in software stocks. According to CNN, industry plugins for Anthropic's Claude Cowork rattled the sector in early February, and a software ETF fell almost 6% in a single day.

A debate that keeps coming back

Over the past few months, I keep seeing the same debate come up.

If OpenAI, Anthropic and Google are investing hundreds of billions in foundation models, why do we still need vertical software?

The thesis is simple. Ever more powerful models will commoditize business logic, eliminate interfaces and erode differentiation. In the end, everything becomes a prompt.

I understand the provocation, but it starts from a flawed premise. Enterprise software has always been about process, and never about code.

Process engineering is the real differentiator

From the outside, every industry looks simple. Finance teams analyze deals. Law firms produce legal opinions. Engineering firms develop projects. Sales teams manage pipeline.

But zoom in. Every organization operates differently. Every team structures its work around its own rules. Every leader has specific quality criteria.

And it is at this level of detail that real value shows up. The last 10% of the process is where 90% of the competitive advantage lives.

The last mile is operating culture

Many people confuse the last mile with surface-level customization. But the last mile is really about embedding operating culture inside the system.

Software is not a neutral tool. It is an encoded opinion about how work should happen. It embeds decision patterns, approval criteria, governance flows and the metrics that actually matter.

When that works, software becomes institutional memory. And institutional memory is something a generic model cannot replace.

A powerful model does not guarantee operational reliability

Models are evolving at an impressive pace. But capability is not the same as reliability.

In critical industries, 90% right is the same as 100% wrong. One detail out of standard invalidates the entire piece of work. A subtle error can cost millions.

The challenge was never just generating an answer. The challenge is knowing when to trust, when to verify, which data to use, in what sequence and in what format each role inside the organization needs it.

That layer is deeply vertical.

Strategic neutrality is an advantage

Another critical point. Depending on a single model is a strategic risk.

Companies that want to build durable infrastructure need to be model-agnostic. The goal is to use the best model for each task, automatically, and not the cheapest one.

The best model for text generation may not be the best for quantitative analysis. The best one for code may not be the best for structured reasoning.

Smart vertical software doesn't bet on one model. It builds the layer that orchestrates all of them.

Where precision is worth the most

There are industries where precision has an extremely high marginal value. They are the ones where an error costs millions, a regulatory failure triggers sanctions or a poorly structured decision compromises strategy.

In those industries, the willingness to pay for reliability grows exponentially. And reliability comes from a deep understanding of the process, and not just from the power of the model.

The right question is who masters institutional engineering

The right question is who will master institutional engineering inside each industry.

Because that is where the real moat is born. Not in the code. Not in the model. In the integration of process, culture, data and orchestration.

The next generation of winning companies won't just be AI-native. They will be process-native.

And that requires closeness to the customer. Team by team. Workflow by workflow.

It's harder. Less glamorous. But that is how you build something that lasts.

In short

• Enterprise software has always been about process, and never about code.
• The last 10% of the process is where 90% of the competitive advantage lives.
• The next generation of winning companies won't just be AI-native. They will be process-native.

What has changed since February 20, 2026 · Editor's note
Throughout 2026, major platforms began packaging instructions, tools and agents for specific tasks. On September 10, OpenAI opened the Agents API in beta, with the same harness as Codex. On September 15, TypeSafe AI unveiled Jev, a model that returns typed decisions with a calibrated confidence level. On September 30, Microsoft swapped agents for plugins in Copilot, with IT approval in a central registry.

Further reading

• From adoption to operations with AI, on portability above the models
• OpenAI opens up the Agents API
• Jev, TypeSafe's decision model

Originally published on LinkedIn on February 20, 2026. Read the original article. This version has been revised and expanded with context and sources. The views are those of the author, a co-founder of OperOS AI.