Five Years After He Advised My Alexa Win, Curtis Chambers, Uber's First Head of Engineering, Reviewed My 29-Agent AI Operations System. Here's What Happened.

On Friday, April 29, 2026, Curtis Chambers had a flight back to San Francisco at five.

He gave me one hour and forty minutes before the airport.

That was enough time to change how I understand everything I have built.

Who Is Curtis Chambers?

Curtis Chambers was the seventh employee and first Head of Engineering at Uber. He was responsible for engineering as Uber scaled its early operations. He is a co-inventor on the patent for Uber's surge pricing — the algorithm that re-priced ride availability in real time based on demand, a system that had not previously existed in on-demand ride services.

He was not just building software. He was building inside a legal vacuum. The legal system at the time only recognized two categories of work — employee or contractor. Uber needed a third paradigm that did not exist yet. So they invented it. The on-demand labor model that later spread to DoorDash, Instacart, and platforms that followed was built during that period. Curtis was part of the engineering team that made it work at scale.

Before Uber, he helped build Expensify. After Uber, he is doing a PhD at Berkeley and UCSF in computational precision health and co-building a startup with his PhD advisor.

He has been my mentor since 2019. He advised and reviewed my prototype before I won first place at the Global Amazon Alexa Skills Challenge in 2021. He has been to PowerFuel Damas events. He has seen me speak. He has watched my entire entrepreneurial journey from the beginning.

On Friday, five years after that Alexa win, I showed him everything I have built since.

What Did I Bring to That Table?

This was the first time I have ever shown the full system to anyone. I needed him to see the whole arc — not just the 29 agents, but where they came from and why I built them.

I told him about 2021. When ChatGPT launched, I was already paying attention in a way most people around me were not. I trained 13 people inside my agency on how to use it — immediately, in the first weeks. Not because I had a plan. Because I could see what it was going to do and I did not want to be behind.

I told him about the Meta AI experiment — the custom AI avatar I built inside Meta AI Studio that generated nearly a million messages. 947,000. That was not a product. That was me testing what happened when you gave people access to a personalized AI presence at scale. The answer was: they came back. Obsessively.

I told him about the 16 custom ChatGPTs I built and put into production inside my agency. Not demos. Not experiments sitting in a folder. Tools my team used. Proposal cycles compressed from 10 days to 3. An AI-native hiring pipeline. Systems that ran whether I was in the room or not.

I told him about my father. About why any of this matters. About what drove me to care about AI before it was a career move or a trend — back when it was just the thing I could not stop thinking about.

And then I showed him what I built in the last 30 days.

What Was on the Table?

I had the architecture map printed and laid flat on the table — because the laptop was too much for a conversation this dense. You cannot have a real conversation about a system while someone is clicking through it. You need to be able to point.

The map showed the full 29-agent orchestration system. Each agent is a persistent AI specialist with its own defined role, its own instruction set, its own domain, and its own memory. They do not overlap. They do not repeat each other's work. They hand off tasks through a central routing hub — agent-pm — the same way a chief of staff routes work through a team.

I walked him through three layers:

• →
  The orchestration layer. 29 agents in Claude Code — agents for SEO, content, brand strategy, research, quality assurance, operations, video, design, coaching, and more. Each with governance files that define what it can and cannot do. A dashboard. A daily reporting system. Context management across sessions.
• →
  The project management layer. A system where humans and agents are assigned tasks side by side, in the same dashboard. Agent-pm routes work. Agents report completions. The whole system is observable. I have not seen this built publicly yet — a hybrid team dashboard where AI agents and human owners appear in the same task queue.
• →
  Three years of receipts. 16 custom GPTs in production. A Meta AI avatar with 947,000 messages. Proposal cycles from 10 days to 3. An AI-native hiring pipeline. The website I built with the system at daniellevantini.com. Applied AI operations running a real agency, not a side project.

What Did He Say?

Halfway through the conversation, I said something I had been working toward for a long time:

"I consider myself an orchestrator because I am orchestrating all these agents, and I am a specialist in applied operations. In AI, at least in the fields of things I have done. But I am still trying to understand what to call it."

Curtis listened. Then he said this:

He kept going. He told me the role I have built is new. Not new like "a new job title." New like: this category of work does not exist at scale yet in business. Most of what people are calling "AI" is a single agent doing a single task. That is not what I built. What I built is an orchestration layer running operations across an entire business — and that is something else.

I will be sitting with that sentence for a long time.

What Is the Name for What I Do?

For years, I did not have outside language for any of this. I just had the work. I knew what I was doing inside my agency. I could describe the system. I could not name the role.

Now I can.

Applied AI Operations. I am a multi-agent orchestrator running a hybrid AI stack — Claude Code for the orchestration layer, and ChatGPT, Gemini, Seedance, and whatever ranks best in the moment for the work underneath. Models and tools are changing faster than most operators can keep up with. Adapting the stack is the work.

I am not a developer. Not an engineer. Not a vibe coder. An orchestrator who knows the business cold and runs it with agents.

The conversation pointed me toward a direction — coaching, speaking, and Applied AI Operations work for non-technical founders and orchestrators. From someone who has watched my entire arc since 2019, that was the part I will carry the longest.

What Is the Legal Vacuum Parallel?

The conversation went somewhere I did not expect.

Curtis drew the parallel between what is happening in AI right now and what happened in ride-sharing fifteen years ago. AI is operating in the same kind of legal vacuum that ride-sharing operated in when Uber launched. The law had not caught up. The categories did not exist. The systems were real but the language to describe them — legally, professionally, commercially — was being invented in real time.

He watched that vacuum close from the inside. He knows what it looks like at the beginning, before the definitions arrive. My read of the conversation: we are there again. The role of manager of agents does not have a job description, a salary band, a LinkedIn category, or a university program yet. That does not mean it is not real. It means it is early.

Someone who was inside the last paradigm shift — who helped build the engineering foundation of on-demand platforms — is exactly the kind of mentor you want when the next one is the paradigm you are building inside.

Some of what he told me, I cannot share publicly. That is the nature of frontier intelligence — you protect the room. But the conversation opened possibilities I had not considered — around deployment models, around how to think about the business and curriculum architecture for teaching this to non-technical orchestrators.

What Does This Mean If You Are Not Technical?

Here is the thing about Curtis's validation that matters most to non-technical founders reading this:

He was not impressed because I wrote code. I did not write code. He was not impressed because I have a computer science background. I do not have one.

He was impressed because I understood what I was building well enough to design it. Because I had three years of operational receipts, not just a prototype. Because I knew the business problem before I picked the tool. Because I wrote the governance — the rules, the constraints, the handoffs — that made 29 agents behave like a system instead of a pile of prompts.

The orchestration layer of AI operations is not going to be built by engineers who do not understand the business. It is going to be built by people who understand the business deeply enough to translate it into systems — whether they write the code themselves or not.

That is the opening. That is where non-technical founders can compete at the frontier. Not by learning to code. By learning to orchestrate.

---

Thank you, Curtis. For 2019. For 2021. For Friday.

I've been here. You just didn't know.