After talking with a few friends across different companies, my FOMO around harnesses and loop engineering faded quite a bit.

What surprised me wasn’t that everyone had built amazing AI systems. It was that almost everyone is still figuring things out.

The harnesses people share online are useful. They can improve general coding workflows and help individuals become more productive. But the conversations made me realize that the highest-value harnesses in production are rarely generic. They’re often tightly coupled with a company’s domain knowledge and internal workflows.

More importantly, I realized that harnesses are just one small part of a much bigger story: how different companies are adopting AI.

The pace varies significantly.

Big tech companies often carry years of technical debt, making large-scale AI adoption harder. AI-native companies are experimenting with much more aggressive AI-first workflows. Meanwhile, industries like fintech and biotech tend to move more cautiously because of strict regulations, high reliability requirements, and domain-specific constraints.

Different industries are moving at different speeds, but they’re all moving in the same direction.

That reminded me of the Industrial Revolution.

We’re not at the “Ford assembly line” stage yet, where software development is fully automated from end to end. Instead, we’re much closer to the early steam engine era. Every engineer now has a far more capable machine at their side, but the work is still largely directed by humans.

The real transformation isn’t replacing people overnight. It’s gradually turning individual craftsmanship into repeatable systems, one workflow at a time.

Today’s AI workflows still rely heavily on human judgment. Developers decide when to trust the model, how to evaluate outputs, when to intervene, and how to integrate AI into existing processes. Over time, the best of these practices will become standardized workflows, then organizational systems, and eventually parts of the software development pipeline itself.

That’s why I don’t think engineers need to panic about AI taking away their jobs.

Every major technological shift changes the nature of work. Some tasks disappear, but new skills, new roles, and entirely new ways of creating value emerge. History has repeated this pattern through every industrial revolution, and there’s no reason to believe AI will be different.

AI lowers the cost of execution, but domain knowledge determines what should be executed. Knowing what to build, why it matters, and how to integrate it into a real business remains the hardest problem.