Edge AI vs. Core AI: Where the Real Value Lives

The majority of companies I walk into right now have employees using AI. That's not the question anymore. The question is whether any of it is showing up on the income statement.

There's a piece making the rounds from Andreessen Horowitz — George Sivulka's "Institutional AI vs. Individual AI" — that frames this problem well. He draws a parallel to the electrification of textile mills in the 1890s. Factories swapped steam engines for electric motors and saw almost no productivity gains for thirty years. The breakthrough came only when they redesigned the entire factory floor around the new technology.

That's exactly what I'm seeing in lower middle market companies today. The technology is in the building. The productivity gains are not.

At Fractional Agent, we use a framework to help clients see the difference clearly. We call the two types of AI usage Edge AI and Core AI. Understanding the distinction — and making a deliberate choice about where to invest — is the difference between AI that feels productive and AI that actually moves your bottom line.

Edge AI: Where Everyone Starts

Edge AI is what most people think of when they hear "AI at work." It's the individual productivity layer — your marketing director using ChatGPT to draft emails faster, your controller asking Claude to summarize a report, your sales team generating presentation slides in minutes instead of hours.

These tools are good. I use them myself. They make individuals better at their day-to-day work, and they're a perfectly reasonable place for a company to start its AI journey.

But here's the problem I keep running into: Edge AI improvements are almost impossible to roll up into measurable financial impact.

If your accounts receivable clerk drafts collection emails 40% faster, that's genuinely helpful for that person. But can you point to a line item on your P&L that moved? Can you quantify exactly how much EBITDA that 40% improvement created? In my experience, the answer is almost always no.

Sivulka puts it sharply: most publicized AI use today is individuals "productivity-maxxing" with zero real impact on firm-level value. He's writing about Fortune 500 companies, but the dynamic is identical in a $20M manufacturer or a $50M distribution business. People feel more productive. The financials don't change.

That's not a criticism of Edge AI. It's a recognition that Edge AI is a starting point, not a destination.

Core AI: Where the EBITDA Lives

Core AI is something fundamentally different. It's AI embedded in your operational workflows — the repeatable input/output cycles that actually run your business. Production scheduling. Inventory management. Demand forecasting. Accounts payable processing. Customer service routing. Quality control.

The distinction matters because Core AI automates the processes where you can draw a direct line from the improvement to a financial outcome. When an AI agent handles 80% of your invoice matching volume, you can measure exactly how many hours that freed up, what those hours cost, and what your team did with the reclaimed capacity. That's a number you can point to on your income statement.

In my career — from running turnarounds at manufacturing companies to integrating acquisitions in distribution — I've learned that the gains that actually change a business are never the glamorous ones. They're the boring, repetitive, high-volume processes that consume hundreds of hours every month. AI that targets those processes creates compounding returns because the savings show up every single cycle: every invoice run, every production schedule, every customer interaction.

This is Sivulka's "redesigning the factory" applied to a $30M family business. You're not just giving individuals a faster motor. You're rearchitecting how work flows through the organization.

Why Most Companies Get Stuck at the Edge

It's not because leadership doesn't understand the opportunity. Most of the owners and executives I talk to intuitively grasp that workflow automation is where the real value is. The problem is that Core AI requires sustained human attention to work — and that's exactly the resource most companies don't have.

Edge AI is self-service by design. You hand an employee a ChatGPT login and they figure it out. Core AI is not. It requires someone who understands the business process end to end, can configure and monitor the AI agents doing the work, validates outputs against business context, and refines the system continuously as conditions change.

The MIT/RSM research that anchors our practice tells the story: AI implementations show strong gains in the first three months, but performance degrades steadily after the consultants leave. By month twelve, most pilots have stalled. By month eighteen, leadership concludes that "AI doesn't work" — when the real problem was that nobody was paying attention.

This is the 95% failure rate we talk about constantly. It's not a technology failure. It's a stewardship failure. And it hits Core AI hardest, because Core AI systems are the ones that need the most ongoing human oversight to keep delivering results.

A Practical Framework

Most companies we work with have plenty of Edge AI activity happening organically. What they don't have is a deliberate strategy for Core AI — and that's where the money is.

Both Matter — But in Different Ways

I want to be clear: this is not an argument against Edge AI. Sivulka makes the point well — individual AI tools are the vector by which most companies first experience the transformative potential of AI. That initial spark matters. It builds familiarity, reduces fear, and creates internal champions who push for deeper adoption.

In fact, Edge AI can be a useful bridge to Core AI. When a plant manager uses ChatGPT to analyze production data and realizes how much faster she gets to an insight, that's a moment of recognition. She starts asking: what if this wasn't me running a one-off query, but an automated process running every shift?

That question is the transition from Edge to Core. And it's our job to help clients make it.

But the transition doesn't happen by itself. It requires someone who understands both the technology and the business well enough to identify the right workflows, design the automation, deploy it, and — critically — keep it running. That's the Fractional AI Agent Manager role we've built our entire practice around.

The Change Management Reality

One thing the a16z piece touches on that I want to underscore from my own experience: the enablement problem is real and it's underestimated.

I've spent most of my career in change management — running turnarounds at manufacturers, integrating acquisitions, restructuring distribution businesses. The pattern is always the same: the technical solution is 30% of the work. Getting the organization to actually adopt and sustain the change is the other 70%.

Core AI deployments require real change management. You're not asking someone to use a new tool — you're asking them to change how work gets done. That takes trust, communication, and someone on the ground who can speak both the language of AI and the language of the plant floor.

So What Should You Do?

Encourage Edge AI. Let your people experiment with ChatGPT, Claude, and whatever other tools they find useful. Don't over-govern it. The familiarity they build becomes the foundation for everything that follows.

But don't confuse it with strategy. Individual productivity gains are real but they don't compound, they're hard to measure, and they won't show up on your bottom line. If your AI story stops at "our people use ChatGPT," you don't have an AI story.

Identify your Core AI opportunities. Pick the three to five operational workflows that consume the most time, generate the most errors, or create the biggest bottlenecks. Those are your targets. The boring work — AP, AR, scheduling, document processing, customer communications — is almost always where the biggest EBITDA impact hides.

Plan for stewardship from day one. The lesson of the 95% failure rate isn't that AI doesn't work. It's that AI without ongoing human attention doesn't work. Whether you build that capability internally or bring in a fractional resource, someone needs to own the AI operations. Not as a side project. As a defined role.

For the companies Jon and I work with every day, that redesign doesn't start with a massive technology investment. It starts with a clear-eyed look at where your business actually runs — the operational workflows that drive your costs, your revenue, and your margins — and a plan to make them smarter.

Edge AI will make your people better at their jobs. Core AI will make your business more valuable. Both matter. But if you have to choose where to invest your next dollar, invest it in the core.

That's where the EBITDA lives.