Scaling Operational Intelligence: An AI First Company

Introduction

The Neural Network of a Company

I often say that operations is the neural network and the intelligence of a company—the way systems and processes use capital to live and grow as efficiently as possible into maturity. Accounting, finance, compliance, risk, supply chain, people management: these functions form the backbone of how any company actually runs. And yet, in my experience, they are often treated as overhead rather than strategic advantage.

For a startup, the balance between operational investment and innovation is delicate. Every hour spent on compliance paperwork is an hour away from customers. The traditional approach—hiring specialists, outsourcing to consultants, purchasing enterprise software—creates costs that grow alongside the company, sometimes faster than the value they provide. I have seen many founders spend more energy learning QuickBooks than learning their customers.

A 3-year audit of our financials and taxes this year took me 20 hours. The traditional approach would have been over 100 hours plus consultant fees.

This is the story of how we restructured our entire operational infrastructure using Somata, reducing complex business tasks to the bare minimum of human interaction while maintaining the quality and compliance standards required of a federally-designated small business operating in highly regulated industries—healthcare and government, two of the most documentation-heavy sectors you can choose.

Somata and Operations

Structuring Complexity into Cells

In our Somata structure, I organized every part of the company into core components and then structured the data into what I call "cells"—highly utilized data subsets that contain everything the AI needs to know about a single domain, whether that's accounting, compliance, revenue, or any other function. The goal was to create self-contained units of context that the model could access independently without needing to pull in unrelated information.

Then came the dialectic learning loop. I would give the model a task, review what came back, adjust the cell structure based on what was missing or inefficient, and repeat. Over the course of the next month, something interesting happened: the task half-life started collapsing. Tasks that once took hours became tasks that took minutes. As the body of context built in the model grew richer, my own understanding of how to structure data for AI sharpened in parallel. By the end of that month, I was functioning as a one-person operations team for a 7-figure company.

Two recent examples illustrate this well. We had our annual taxes and a program-wide audit of the 8(a) program due at the same time, and I also wanted to move away from outsourced accounting software. The problem with platforms like QuickBooks isn't the software itself—it's the dependency they create. Once you're on that platform, you spend enormous energy learning their language, their data model, their quirks. How many QuickBooks experts have you met in your career? All that expertise represents time invested in someone else's system rather than your own.

I would rather own the data and the dependency myself. When you structure your own data for AI, you are building something that compounds over time—every refinement makes the next task faster, and that learning stays with you. The alternative is molding yourself to external software, where all that accumulated knowledge never transfers when you eventually move on or the platform changes beneath you.

So I examined our accounting structure and mapped core data to core tasks. The Accounting Somata has a Reports cell containing everything needed for both taxes and audits: trial balances, profit and loss statements, bank statements, all integrated with Mercury, our banking platform. From there, I extended this approach to other cells—revenue, expense, debt—each one a self-contained data structure that the AI could query independently without pulling in irrelevant context.

The result: a 3-year audit of our financials and taxes took me a total of 20 hours, and most of that time was spent architecting the data structure in the first place. The actual audit work—the part that would normally consume weeks of back-and-forth with accountants—was reduced to simple queries against well-organized cells.

This task life decay has real compounding effects on the business. We work in compliance-heavy industries—healthcare and government—where the rules are complex and constantly changing. I built the same cell structure for compliance that I had for accounting: contract analysis, agreement review, regulatory tracking. The same pattern produced the same acceleration.

The net result is a fundamental shift in how we use external providers. Instead of outsourcing 100% of a function, we now outsource only 5-10%—and it's the right 5-10%, the parts that actually require specialized expertise. The strategic decisions, the final reviews, the judgment calls that demand someone who has spent twenty years in the field. We can afford to hire the best because we're only paying for their genuine expertise, not their time on commodity work. At the other end, the truly manual tasks—being on calls with benefit providers, filing state certifications—can be done at low cost. The middle 80-90% that used to consume most of the budget? Somata handles it.

The overall effect is operational foundations that are stronger, leaner, and faster than those of companies many times our size. We compete in contracting and legal realms against organizations with ten times our headcount, and we win. But operations is only half the equation—the other half is the lifeblood of every company: revenue.

Alignment Driven Business

Accelerating Directional Revenue

If operations is the neural network, then cash flow is the lifeblood—the thing that keeps the whole system alive long enough to learn anything at all. A sustaining company does not exist without it. And there are many ways to chase it: contracts in a given industry, partnerships, incubators, series funding. Early-stage companies know all of these paths intimately, because survival demands fluency in them.

I often say that capital is an accelerant, not a direction. Money will get you wherever you are pointing faster—which is exactly the problem if you haven't aligned the trajectory first. Without that alignment, you don't reach escape velocity; you reach free fall. But there's a productive tension here, too. Sometimes the pursuit of capital itself reveals where you were always meant to be. The pressure clarifies. So the question becomes: how do you chase revenue in a way that also teaches you who you are?

Before I started architecting the data infrastructure, I wanted to solve a more fundamental problem: how do we identify the customers who will take us where we actually want to go? Not just any revenue—directional revenue. Engagements that compound into something, that build toward the kind of company Somata is meant to become. I started coalescing variables—time in industry, project scope, expected duration, AI readiness, revenue potential—into what I called an alignment score. The score wasn't static; the weights could shift as the business evolved, emphasizing different factors at different stages. And critically, this was exactly the kind of judgment that the model could eventually learn to make alongside us—another application of the trust-the-model principle that became our operating KPI.

To operationalize this, I built a tool using Claude that integrates all our capture channels—emails, direct company sign-ups, manual entries—and performs baseline research on each incoming lead. The model pulls publicly available information: the company's vision statements, brand positioning, ethos, public filings if they're traded, recent press releases, leadership commentary. All of this gets categorized into component scores. AI Readiness, for instance, breaks down into sub-variables: frequency of AI-related announcements, explicit vision statements about technology adoption, what leadership is saying publicly about digital transformation. Each of these rolls up into its piece of the overall alignment score.

The result is an interface that surfaces which leads are most aligned with where we want to go as a company—not just who has the biggest budget, but who will take us somewhere worth going.

The system is still evolving—we're working with publicly available data, which limits precision—but the vision is clear. I think often about Bloomberg terminals: those dense, matrixed interfaces that dominate trading floors, with dozens of data streams tracing a company's press footprint, supply chain relationships, financial movements, partner networks, even mineral and material exposures. That's the aspiration. Not a CRM that tracks emails, but an intelligence layer that understands the shape of a potential relationship before the first conversation happens.

2025 Outlook

The Compounding Effect

To give you a sense of how dramatically this compounds, here's a snapshot of what we're projecting for 2025. These aren't theoretical savings—they're the result of restructuring operations through the Somata model, replacing fractional external headcount with AI-augmented internal capability:

The Modern AI Company

The Creators Economy

I sometimes wonder what the future of economics looks like when human capital is no longer the primary constraint. Our economic models have always been built around it—hence the obsession with metrics like revenue per employee, production levers, headcount efficiency. But in an agentic AI landscape, where more tasks are executed through AI with less friction, those measures start to break down. We're already seeing the early signs: the race toward energy infrastructure, the erosion of entry-level positions, the compression of entire job categories. All of which, understandably, causes real fear.

But here's the tension I keep returning to: how many smart, creative people are trapped in the current economy? People with tremendous ideas who are burned out from the work required just to afford a baseline of living. People who could never bring their visions to life because they lacked the technical skills, the design expertise, the capital to hire those who did. What happens when those barriers disappear?

I think we're on the edge of something I call the creator economy—not the Instagram version, but something more fundamental. A shift in who holds the power to build. Consider our current relationship with software: we pay for services that meet maybe 40% of our actual needs, but it's enough that we tolerate it because the alternative is building our own from scratch. That's been prohibitive. But in a world where you can build applications and services individually tailored to your exact requirements with increasingly minimal effort—why would you pay for something that meets only a fraction of what you need? The Overton window is shifting. People are starting to realize how much can be done with where we even are now.

My prediction—and my bet—is that AI-first companies like ours will take increasing market share in the years ahead. Companies that are AI-native from the beginning, operationalized to scale into this emerging landscape, have very little reason to slow down. They'll continue moving through dialectic loops like the Somata model: accelerating how they interact with AI, identifying what can be agentified, compressing timelines further, and repeating. The compounding is relentless.

As an emergent trend, I think the future of operations looks something like this:

• The core function becomes designing and managing agents, data structures, and dialectics with AI models—making the act of running a company a fraction of a person's day. I haven't hired anyone yet, and it's not because I don't want to grow. It's because task decay keeps outpacing the need for additional headcount. When I do make my first hire, it will likely be a Director of Agentic Operations—someone whose job is to move through as many of these learning loops as possible, as fast as possible.
• Small teams with outsized revenue will become the norm. The traditional playbook—raise money, add people, grow—doesn't map onto a world where the work those people would have done can be compressed into an hour of someone's day. I think multi-million dollar revenue per employee will become standard, not exceptional.
• The savings from operational efficiency will be reinvested into highly intelligent people—people whose salaries will likely balloon, because their value lies in codifying their problem-solving and thinking into both models and output. The premium will be on judgment, not execution.

We're already well on our way.

Beyond the direct dollar savings, this level of operational efficiency eliminates an entire category of hidden costs that comes with scaling: the overhead of people management as teams grow, the need for management-layer hires, the friction of talent acquisition and onboarding. These costs don't show up in obvious line items, but they compound relentlessly in traditional organizations.

The leverage here is difficult to overstate. Each one of us wears multiple hats that would traditionally require dedicated headcount—roles that, in a conventional organization, would mean hiring specialists, building teams, managing the complexity that comes with coordination at scale. The Somata structure doesn't just save money. It makes it possible for three people to operate with the footprint of a much larger organization.

This is what an AI-first company looks like in practice: not replacing humans, but amplifying each person's capacity to the point where traditional organizational structures become optional rather than necessary. The question isn't whether this model works—we're living proof that it does. The question is how far it scales, and what happens to the broader economy when more companies start to figure this out.