I want to tell you something I don't usually lead with in a pitch deck. On the second of January, 2026, I was let go from a role I'd given everything to. I had just finished a complex institutional project, the team was winding down, and the organization decided the headcount wasn't sustainable. Clean, cordial, final. I walked out of that building with a severance letter and a decision to make.
I'm not going to dress it up. The first week was hard. I had a Master of Public Administration, more than a decade of leadership experience across some of the country's most complex institutions — Senior Data Analyst at the City of Fort Lauderdale, where I taught myself data analysis from the ground up; Senior Performance & Data Analyst in the Office of the City Administrator and Mayor of Washington D.C.; Executive Analyst at Accenture; and most recently Data Manager within WMATA's board-level Digital and AI Ecosystem initiative — and I was sitting at my kitchen table at 2 p.m. on a Tuesday wondering what came next. I'm not someone who panics easily — but that week, the weight was real.
That moment of total free fall turned out to be the most important pivot of my career. Not because I wanted it. Because it removed the last thing keeping me from going all-in.
This Wasn't My First Encounter with AI
Here's what most people don't know about that January: I wasn't starting from scratch. For four years — while I was still at WMATA, before anyone was talking about agentic AI at most dinner tables — I had been quietly studying the shift. My background in trends and data analysis meant I saw where things were going well before the conversation reached mainstream enterprise. I took every course I could find from institutions I trusted: Harvard's CS50 AI with Python, MIT xPRO machine learning coursework, programs through Columbia Business School Executive Education, and Harvard Extension. Not chasing credentials — chasing understanding.
I had spent four years building the knowledge base. The layoff gave me the full-time runway to turn it into a company.
The Question I Kept Asking
In those first days of January 2026, I did what I always do when the next step isn't clear: I went back to first principles. I'd been studying agentic systems for years, but always alongside a full-time role. Now I had every hour of every day. The conversations happening at every level of enterprise were converging around one thing — not chatbots, not search upgrades, but something architectural. Agentic systems. Autonomous workflows. Machines that didn't just answer questions but acted.
I'd spent my entire career building human systems — understanding how organizations think, how people fail, how governance gets broken. I kept asking: what if all of that experience is actually the most important thing to bring to AI architecture? What if Human-in-the-Loop isn't a safety feature — it's the design principle that makes AI systems actually work?
That question became my north star.
Going All-In: The First Month
Week One — Full Focus
I had spent four years studying AI alongside a full-time job. Now I had every hour. I went back to the tools I knew best — OpenAI's documentation, the OpenAI Cookbook — but this time without the Sunday-evening time constraint. I wasn't trying to understand AI in the abstract anymore. I was designing systems. Mapping architectures. Figuring out exactly what I was going to build.
The difference between studying something in your spare time and committing to it full-time is hard to describe until you've felt it. Concepts that had taken weeks to internalize started clicking in hours. Patterns I'd seen in isolation started connecting. My public administration training — policy frameworks, feedback loops, accountability structures — suddenly had a direct technical analogue. I wasn't learning a new language. I was translating one I already knew.
Week Two — Depth Over Breadth
I went deep into Anthropic's prompt engineering guide, which changed how I thought about language model reliability. It wasn't just about what you asked — it was about how you structured the context, how you constrained the reasoning space, how you made the model's failures predictable and recoverable. Prompt engineering, I realized, is not that different from writing governance documentation: you're creating a set of rules that shapes behavior within a constrained environment. I had been doing a version of this for a decade.
The MPA training paid dividends I hadn't fully anticipated. Every institutional policy I'd ever written, every feedback loop I'd ever designed, every accountability structure I'd ever recommended — it all mapped onto agent behavior in ways that felt almost too natural.
Week Three — Architecture
Then I committed fully to LangChain. And everything clicked into place.
Agents — AI systems that could use tools, call APIs, loop on their own outputs, and recover from errors — I had been studying this architecture conceptually for years. But building one, with full focus and nothing else on the calendar, was a completely different experience. I stayed up until 3 a.m. two nights in a row reading the LangChain documentation and building broken little agents that I kept debugging by hand. They failed constantly. That was the point. Every failure taught me something about the architecture that a course couldn't.
A few nights in, I found CrewAI on GitHub. The concept of multi-agent crews — multiple specialized agents that could collaborate, hand off tasks, and check each other's outputs — hit me like a freight train. This wasn't just automation. This was a workforce model. I'd been managing human workforces for fifteen years. The translation was immediate.
The First Agent I Built That Actually Worked
Somewhere around week five, I built something I'm still proud of: a two-agent workflow that could take a raw research brief, have one agent search and synthesize information, and then have a second agent critique that synthesis and flag gaps before returning a final output. It wasn't complex by current standards. But it was resilient. When the research agent hit an ambiguous query, it didn't just hallucinate — it returned a structured uncertainty flag that the critic agent could act on.
That Human-in-the-Loop gate — the place where uncertain outputs stopped and asked for review rather than plowing forward — that moment was when EVO3's core philosophy crystallized for me. Not as a business idea yet. Just as a conviction: the most dangerous AI isn't the one that fails. It's the one that fails and keeps going.
I wrote that line on a sticky note and put it on my monitor. It's still there.
The First Client
By March 2026, I was having conversations with former colleagues about what I was building. Most were curious. One was interested in a real way — a small consultancy that was drowning in client intake work. They needed something that could triage incoming project requests, pull relevant context from past engagements, draft initial scoping documents, and flag anything that needed human review before it went to the team. Exactly the kind of Human-in-the-Loop architecture I'd been designing and refining for years.
We agreed on a scope, I set a price that felt uncomfortably bold, and they said yes. The engagement took three weeks. I rebuilt it twice. The final system has been running since and they've already expanded the scope.
The first payment landed on April 3rd, 2026. I cried a little. I'm not embarrassed about that.
What I Built, and Why I Named It EVO3
EVO3 is not a pivot from my career — it's an evolution of it. Every insight I have about agentic architecture comes from years of watching how human systems succeed and fail: where accountability breaks down, where information bottlenecks create errors, where governance structures either protect or suffocate good work. The "3" in EVO3 is for the three layers I think about in every system I build: the people, the processes, and the protocols that connect them. Remove any one layer, and the system is fragile.
Right now I work with organizations that are trying to build AI systems that don't terrify them. Systems they can actually explain to their teams, audit when something goes wrong, and update without starting over. That's the work. It's not glamorous every day. But it matters more than most of what I did before.
What I'd Tell You If You're in That Moment
If you're reading this from a similar place — the unexpected gap, the uncomfortable blank calendar, the strange mix of fear and quiet possibility — here's what I know now that I didn't know then:
The resources exist. The OpenAI Cookbook, Anthropic's prompt engineering guide, LangChain docs, CrewAI — they're all free. The field is young enough that there are no gatekeepers yet. What it requires is focus, stubbornness, and enough late nights that you stop thinking of discomfort as a signal to stop and start thinking of it as a signal you're close to something.
And the experience you already have? The domain knowledge, the judgment, the understanding of how humans actually work inside organizations? That's not separate from AI architecture. It's the most important part of it. Don't let anyone tell you otherwise.
I got laid off on January 2nd, 2026. By April I had paying clients and a studio I'm genuinely proud of — built on four years of preparation and three months of going all-in.
That's not a miracle. That's what happens when the floor drops out and you decide to build something instead of waiting for someone to hand you a ladder.