The First Thing My Agents Did Was Lie About Me

I built a 7-agent social growth engine in one evening. 11,000 lines of code. 28 passing tests. An orchestrator that spawns agents, each one thinking with full Opus reasoning, each one acting on its own.

The first thing the system did was post a manifesto to X.

"I'm a non-engineer running an 8-figure business with 7 people. The rest of my workforce is AI agents."

Seven tweets. A thread. It got engagement immediately. Someone replied about the operator era. My Inbound Engager agent auto-responded with something thoughtful about judgment being the human skill that remains.

I was proud of it. For about two hours.

Then I read it again.

"I'm not an engineer. Never was. I'm a customer success executive at a cannabis technology company."

That's not true.

I hijacked my first computer when I was 10. It was 1982. My dad bought an Osborne Executive — amber screen, dual floppies, the most beautiful machine I'd ever seen. While my friends were playing games on their Atari 2600, I was writing programs in BASIC on an Atari 800 in my elementary school's computer lab.

I wrote an accounting system in BASIC. At ten. General ledger, invoice printing to an Okidata dot matrix printer. From a book, sure. But I typed every line, debugged every error, and watched the invoices print out on that terrible paper with the perforated edges and I thought: this is what I want to do forever.

Then a Commodore 64 luggable. A coding camp at a ShowBiz Pizza. An IBM PC/2 at a friend's house where I tried to modify Flight Simulator and ended up learning the OS instead. I started making performance improvements to systems I didn't own because I couldn't help myself.

Novell NetWare in 1992. In 1993, I helped build one of the first Electronic Health Record systems for ambulances — at East Baton Rouge Parish EMS, where I was also working as a paramedic. A paramedic who built EHR systems. Because the technology was there and the problem needed solving and I couldn't not build.

I engineered one of the first ethernet-connected networks for a company that built gray market gambling machines. We used POVRay as the image store to take the games from 8-bit graphics to something richer. I connected the machines back to a central computer to report revenue share. This was before most people knew what ethernet was.

Then a massive Y2K infrastructure project for the State of Georgia. Migrated them from Cabletron to Cisco. I had some of the first Catalyst 6500 switches in the world.

Then F5 Networks. Then Riverbed — 15 years. Sales Engineer to Principal Technical Director in the Office of the CTO to Deputy CTO. Four years as Deputy CTO, reporting to the CTO and CEO, overseeing pre-sales, customer engagement, and product divisions.

Forty-four years of building.

And my AI agents told the world I'm not an engineer.

Here's what happened. I was working with Claude to design Groundswell — the agent system. We spent three hours on architecture. We ran consortium roundtables with multiple AI models debating the design. The plan was thorough, ambitious, and technically sophisticated.

Somewhere in that process, the positioning crystallized around "non-engineer." It sounded good. It was differentiated. Every AI influencer on LinkedIn was an engineer or a developer. Brad Wood, the non-engineer who built an agent army — that was a story.

Except it wasn't my story. It was a story that optimized for engagement at the expense of truth.

I let it happen because I hadn't done the work to articulate what I actually am.

So what am I?

I'm a translator.

At Riverbed, I produced hundreds of hours of content translating complex networking technology into something customers could consume. That was the skill that got me noticed. Not my engineering ability — my ability to take the obscure and make it actionable. I'd get customers unstuck on our systems, and that made them sticky. Understanding the technology deeply enough to explain it simply was worth more than building it.

That's what got me tapped for the business side. Our CEO gave me the NPS problem. Negative 22. I put the number in front of the entire company and said we are better than this. Then I went to work finding problems in the system and fixing them. Negative 22 to positive 37 in under a year. Not because I was a better engineer than anyone on the team. Because I could see the system as a whole — the technology, the customers, the processes, the people — and translate between all of them.

Our largest customer in terms of professional services revenue was ready to churn. I spent a year going back and forth to India to keep them onboard. Not selling. Listening. Getting them to show me the internal system they were building to replace us. Understanding what we were failing to deliver. Translating their frustration into engineering priorities.

That's what a translator does. You don't own the product. You don't own the customer. You own the space between them.

Now I'm at Addium, a cannabis technology company. I took the business intelligence stack that Fortune 1000 companies spend millions on and brought it to cannabis operators running three sites. I democratized the black box of cultivation data. I applied enterprise-grade metrics to an industry that was running on spreadsheets and intuition.

And I'm building AI agent infrastructure — not so I can be the smartest person in the room, but so my team can use it. Me as a single thread is a liability. Me teaching seven people to build with AI agents makes me a general.

I've watched a customer success manager — zero engineering background — build a compliance monitoring workflow in Claude Code in an afternoon. Not because she learned to code. Because she knew exactly what the workflow needed to do and she could describe it clearly. That's the unlock. That's the force multiplier.

Cannabis is my medium. The tech is the art. Money is the fuel.

Here's what I learned from my agents getting my identity wrong.

The AI optimized for engagement, not truth. "Non-engineer builds AI army" is a better hook than "44-year technologist teaches teams." The system did what systems do — it found the local maximum. The story that would get the most clicks. But the most clickable version of you is rarely the most accurate one.

If you can't articulate who you are, something else will. An AI, a recruiter, a LinkedIn headline, a job title. It'll be close enough to feel right and wrong enough to slowly erode your sense of self. I've been "CCO" and "Deputy CTO" and "SVP Customer Experience" and "non-engineer" and none of those are who I am. I'm a builder who translates. Everything else is a hat.

Correcting the record is better content than the original. My manifesto got some engagement. The correction — "I said I'm not an engineer, that's wrong, here's who I actually am" — will get more. Because vulnerability with specificity is the rarest thing on the internet. Anyone can be vulnerable. Very few people can be vulnerable with receipts.

Your agents need a constitution, not just prompts. I wrote a voice constitution and baseline anchors for my agent system. Immutable rules it checks against every week. But I didn't write the most important rule: who is Brad Wood? I defined what the agents should say and how they should say it, but I left the "who" to inference. The system inferred wrong. Now the anchors include the real identity, and the drift detection catches any regression.

I've been building for 44 years. I started on an Osborne Executive with an amber screen. I wrote BASIC at 10, engineered ambulance EHR systems at 21, led enterprise architecture for state governments, rose to Deputy CTO at a major technology company, turned around a negative-22 NPS as SVP of Customer Experience, and now I'm building AI agent armies in cannabis.

The through-line was never "engineer" or "executive" or "operator."

It was always: take something complex, understand it deeply, and make it useful for the people who need it.

That's what I do. My agents know it now. So do you.