We’ve been told for two years that the path to AI ROI is paved with bigger, smarter, and more expensive frontier models. We’ve treated “Intelligence” as the product itself.

But after spending time testing agentic workflows—those loops where AI actually does things—I’m starting to see a massive disconnect between how we’re billed and how the work actually gets done.

In December 2025, I published the Decision Authority Placement Model. The core argument was simple: when enterprises automate execution without explicitly placing decision authority, failure isn’t a risk. It’s a structural certainty.

I built it from practice. Not a survey. Not a lit review. From watching the same failure pattern repeat across virtualization, cloud migr...

I set out to understand what was actually happening inside these agent systems.

So I wired up OpenClaw on a local DGX Spark, gave it a real enterprise question, and let it run.

The answer was fine.

The system wasn’t.

That’s where this got interesting.

The first run looked like what you’d expect from the curr...

But when you actually walk into enterprise environments, a pattern is emerging:

Massive, centralized “AI factories.”
Big iron. Dense GPU clusters. Built as if every workload looks like model training at scale.

That’s usually when I get suspicious.

Because most enterprise AI doesn’t look like that at all.

Enterprise AI workloads are:...

I pulled fifteen years of my own data—emails, tweets, conversations. Over 100,000 messages. Tens of thousands of public posts. The raw material of how I think, how I communicate, and how I’ve built a business.

The goal was simple.

I wanted to understand what actually works. What leads to a deal closing? What causes something to stall? What patterns exist that I’ve...