The long version.

Fourteen years.

I came to software early and obsessively. My family farms; they spend their lives doing work that doesn't pay off until the season turns. Patience was the household language. The assumption that the work is the point. I inherited that. The obsession is what I added.

The first production system I shipped taught me the difference a coder can't see and an architect can't unsee: a coder writes what the spec says. An architect reads the spec and asks what the business actually meant to ask. One of those jobs scales. The other one becomes the spec.

Fourteen years. Seventeen certifications. More enterprise Salesforce trenches than most firms see in a decade. Experience Cloud, Health Cloud, Service Cloud, Financial Services Cloud: integration layers that speak six protocols before breakfast. I've written the spec before the JIRA existed. I've been the person the room turns to when the JIRA is wrong.

Somewhere in there I stopped thinking like an engineer and started thinking like a strategist. Most of architecture is politics, not code. Every system is a pile of competing interests, and sooner or later one of them leans on the weakest joint until it gives. The clean answer usually shows up before I can prove it's right. I write the proof anyway.

What those years really bought me is pattern memory. I can sit with a system for ten minutes and tell you what it actually is, where it's headed if nobody steps in, and whether that's a future anyone wanted.

"I stopped shipping features. I started shipping the systems that ship features. That's the whole pivot."

When the first credible wave of autonomous agents landed, most architects reached for it like another tool in the belt. I saw a new delivery team. One that didn't sleep, didn't argue about sprint length, didn't care about office politics. The question wasn't "how do I use this?" The question was: what does an enterprise software engineering practice look like when half of it is synthetic?

That's when the lab opened. Not to chase hype. To chase leverage. Off-grid, on my own time, out of pocket. v1 to see if agents could run a practice. I'm on v15 of the agent OS, v8 of the convergence layer, v20 of the self-evolving substrate. Each version taught me what the next one had to become.

What I'm building now stays in the lab and unpublished, so I'll be vague on purpose. A substrate that remembers and compounds. Agents that don't deliberate. They dispatch. A convergence layer that pulls scattered signal from everywhere and hands back one coherent answer, fast enough that I honestly shouldn't be getting it out of the hardware I'm running. I hold it to ISV-grade standards and publish it defensively instead of patenting it. The budget is humble. The ambition isn't. I build what most architects only describe in whitepapers.