The Engineer Skill That Outlasts AI Hype: Process Auditing

Dave Ebbelaar draws a clean, useful line through the chaos: companies move from AI-enabled (employees use tools like ChatGPT) to AI-first (processes are re-engineered around AI agents) to AI-native (businesses built from scratch to run on agents). This isn't just taxonomy. It's a roadmap for where the money and demand will flow.

For hands-on engineers, the signal buried in this model is clear: narrow specialization becomes a liability. The developer who only writes backend APIs or only builds dashboards is the bottleneck AI agents bypass. The work that remains — and compounds in value — is end-to-end automation. That means stitching together databases, deployment, frontends, and now agent logic. Ebbelaar's advice to "go full stack" reflects this reality, but with a twist: the stack now includes workflow mapping and business process auditing, not just code.

The hidden skill he emphasizes — identifying automatable workflows inside messy organizations — is where engineers often stumble. Businesses don't have clean process diagrams. They have John, who knows that if the data's not in system A, check system B, and if it's the first of the month, do thing X. Extracting that tacit knowledge and turning it into reliable agent-driven automation is a consulting skill as much as a technical one. Developers who can look at a chaotic process, spot the waste, and say "I can script that in 30 minutes" will capture outsized value.

However, the video leans hard into a freelancer/consultant framing. The core insight — that developers must pair technical execution with process intuition — is valid for employees in larger orgs too, but the "go work for small local businesses" pitch is aspirational. Most technical leaders will find it more practical to apply this thinking internally: refactoring their own team's workflows, not selling AI audits on Main Street. The hype around fully AI-native companies is also just that — hype. We are still in the messy, human-heavy transition, and the real engineering challenge is building agent systems that handle the edge cases John navigates without thinking. That's a systems design problem, not a prompt-engineering one, and it demands strong software fundamentals beneath the AI sheen.