Becoming AI native doesn’t mean buying off-the-shelf products. It doesn’t mean bolting an AI widget onto an existing workflow. It doesn’t even mean getting your team Claude subscriptions. There’s another level to it, and getting there requires thinking from first principles.
The question we start every engagement with: if you were building this business today, from scratch, knowing what AI can do — how would you build it?
Most operators haven’t done this exercise. They’ve mainly thought about AI in terms of “where can I slot it in?” Where might it help speed something up or reduce an error rate? That framing leads to bolt-on solutions. The first principles framing leads to a completely different architecture.
Why the Question Is Hard
If you’ve built a business to scale, you did it one step at a time. The problem at the beginning was getting leads. You solved that, then moved to onboarding. Solved that, moved to service delivery. Hired a manager to run delivery, went back and improved the lead system. Hired more salespeople. Every decision along the way was the best available option at the time.
Over years of making those decisions, you’ve built an operation. Certain inputs produce certain outputs. You have a team executing workflows, management layers coordinating them, and bottlenecks you’ve accepted as the cost of doing business.
“That’s just the nature of our business” — I hear that all the time. The accounting person can only invoice so fast. The project managers can only handle so many clients. The analysts can only process so many deals. These ceilings feel permanent. The only solution has ever been: hire more people.
But most of those ceilings aren’t real constraints. They’re process limitations — artifacts of however the best option played out when the decision was made years ago. Someone built a workflow. The next person followed it. Over time, the tech stack bloated, roles got micro-specialized, and the operation calcified around patterns that made sense once but might not make sense anymore.
AI changes the equation entirely. The constraint of human throughput — the thing that’s always been the bottleneck — is no longer a given. Which means the first-principles question isn’t just a thought experiment. It’s the starting point for a structural transformation.
A Real Example: Real Estate Underwriting
One of our clients is a real estate investment firm with 35+ years of experience. They own 17,000 multifamily units across 16 states. They acquire affordable market-rate apartment communities in secondary markets — not New York or Miami, but Des Moines, Kansas City.
Their number one constraint on the investment side was volume. How many deals they could analyze in a given year.
The process: brokers send deal packages — PDFs, Excel files, operating memoranda, T12 financial statements, rent rolls. An analyst manually opens each document, reads through the financials, and copies the relevant data into the firm’s purchase template — an Excel model with expense categories and formulas that spits out an investment recommendation.
Each deal package takes 25 to 30 minutes. At that pace, they analyze about 250 deals per year. With a roughly 5% hit rate, they invest in about 12 to 13 properties annually, representing around $150 million in deal value.
The bottleneck wasn’t deal quality. It wasn’t capital. It was how fast a human could copy numbers from PDFs into a spreadsheet.
The Bolt-On Approach vs. First Principles
If we’d come in asking “where can we use AI to help?” — the bolt-on framing — we might have built a document processor that speeds up the copy-paste step. Maybe we double throughput. 500 deals a year instead of 250. That’s a great case study. We could have patted ourselves on the back.
Instead, we asked: what does the investment pipeline look like if the capacity to analyze deals is no longer the constraint?
Different question. Completely different outcome.
We built a system that handles the entire underwriting workflow. Upload the deal package, drag and drop. Automated extraction across different document formats. Intelligent categorization of line items based on the firm’s specific logic — because accounting categorization involves judgment calls that vary by firm. Confidence scoring. Context from 35 years of transaction history and corporate knowledge.
That last part matters enormously. One of the biggest reasons AI hasn’t fully taken over underwriting is organizational context. Experienced analysts have decades of pattern recognition — they can sniff a good deal faster, read between the lines on financials, pull from past transactions to evaluate current opportunities. Without that context, an AI system produces generic output that doesn’t earn trust.
We built the context layer in. The firm’s institutional knowledge, transaction history, market intelligence — all feeding the system’s decision-making. Not a generic document processor. A system that reasons the way their best analysts reason, informed by their specific experience.
The Numbers
Before: 250 deals analyzed per year. ~5% hit rate. 12-13 investments. $150 million in annual deal value.
After: capacity for 1,700 deals per year. Same 5% hit rate. Roughly 85 investments. $1.05 billion in potential asset value.
A 7x increase in analytical capacity. From $150 million to over a billion in potential deal value on the balance sheet. All because we removed a bottleneck that was fundamentally about how fast humans could move data between documents.
Do those numbers play out perfectly? We’ll see over time. But the capacity constraint is gone. The conversation shifts from “can we analyze enough deals?” to “what’s the best use of our expanded capability?”
The Next Bottleneck
Here’s the part most people miss. Removing the underwriting bottleneck doesn’t mean the business scales smoothly from here. It means the bottleneck shifts.
If they’re investing in 7x more properties, the property management side — hiring maintenance staff, managing tenants, handling operations across 16 states — becomes the new constraint. More properties, more management complexity, more operational overhead.
Traditional response: hire more people. First-principles response: apply the same AI-native methodology to property management.
This is the pattern. Identify the constraint. Remove it with an AI system built from first principles. Watch where the bottleneck shifts. Apply the same approach to the new constraint. Each cycle compounds — better margins, more capacity, higher revenue per employee.
Constraint → Remove → New constraint → Remove → Repeat.
The companies that execute this pattern across their entire operation, function by function, workflow by workflow, end up with something that feels fundamentally different from a traditional business. Same revenue targets, same market, same customers — completely different operating model.
How to Start This Exercise Yourself
Step one: Write down what your business delivers. Not how — what. The real estate firm acquires high-performing multifamily properties. That’s the output.
Step two: Identify the growth constraint. If you 10x’d your inbound tomorrow, where would it break? If you had unlimited demand, what would prevent you from fulfilling it? Your mind probably goes straight to “we’d have to hire a bunch of people.” That’s the constraint.
Step three: Ask whether the constraint is real or artificial. If the bottleneck is human throughput — the number of people doing a task — that’s an artificial constraint in 2026. It was real when hiring was the only option. It’s not anymore.
Step four: Design from first principles. If this workflow were run by an AI system, what data would it need? What context? What integrations? Where do humans still need to step in? Where are the failure points? Who’s accountable for outcomes?
Step five: Start building. Module by module. Highest leverage first. Measure against the metrics that matter. Feel the ROI. Move to the next bottleneck.
You could do steps one through three in an afternoon. Step four might take a week of focused work. Step five is where you need a team that knows how to build production-grade AI systems. But the thinking — the first-principles redesign of how your business could operate — that starts with you.
The businesses that win this decade won’t be the biggest. They’ll be the ones that asked the right question early: if I were building this today, knowing what AI can do, what would it look like?
And then actually built it.
