Non-AI, AI Companies

For every sufficiently promising technology, companies will claim to be focused on it whether they are or not. It happened with internet companies, social companies, crypto companies. It’s happening with AI right now. That isn’t particularly interesting. What’s interesting is AI is the first of these technologies to change how companies operate.

If you time travelled 26 years into the past (2000), to work at a company, you might snicker at how bad the software was and how slow the hardware was, but how you worked (e-mail, taking notes in meetings, etc.) and how the company operated (individual contributors, layers of management, etc.) would be very familiar.

I suspect if you time travelled 26 years into the future (2052), how you worked and how the company operated would be unrecognizable. A high resolution picture of what this will look like is anyone’s guess, but my guess is the low resolution picture is smaller, flatter companies and many more of them.

I don’t think you’d need to travel that far into the future to see it. Some companies are already operating this way.

This operational shift is already underway, and it’s so meaningful it’s worth arguing for a second dimension of what it means to be an “AI company”:

1. The Product – Is AI the core of the product you’re selling?
2. The Company – Are you using AI to meaningfully operate your business?

The Product

This dimension is the same as in previous technology cycles, so it’s straightforward. You’re either selling (1) the technology itself or (2) a product built on the technology. The model companies (Anthropic, OpenAI, DeepSeek) sell the technology. The application layer (Cursor, Harvey, Perplexity, Sierra) sells products built on it, with varying levels of depth and defensibility.

For the rest of the post, I’ll primarily refer to products built on the technology. The number of companies that sell the technology itself (foundational models) is small, and few new ones get started at this stage.

The Company

This dimension is unique to AI. The metric for whether you’re “using AI to meaningfully operate your business” is whether your ratio of output per employee (revenue, contracts reviewed, properties managed) is significantly higher than competitors in the same industry. Not 20% difference an order of magnitude different. The kind of difference that can only be achieved by being structurally different.

That doesn’t happen by telling everyone at your company to use AI or buying them Claude licenses. Individual adoption is required but not sufficient. What generates a significantly higher difference is building organization-wide infrastructure: shared prompt libraries and agent configurations, business processes owned end-to-end by AI with human oversight, and workflows where AI is the default rather than the exception.

Here are a few recent examples.

1. Shopify – Shopify’s CEO Tobi Lütke mandated “reflexive AI usage” as a baseline expectation. Teams must demonstrate why AI can’t do the work before requesting additional headcount. AI proficiency is part of performance reviews.
2. Crescendo – Crescendo acquired BPO (business process outsourcing) firm PartnerHero and rebuilt operations around AI. They now automate up to 90% of frontline support tickets and run at 60%+ gross margins, roughly 4x a traditional BPO.
3. Anthropic – Anthropic shipped 120 features in 90 days in early 2026, roughly one every 18 hours. Their CPO confirmed that the vast majority of Claude Code is now written by Claude Code itself. The product builds the product.

Looking at these two dimensions together, and since they are independent, we get these four quadrants.

True AI (AI, AI)

These are companies that sell a product built on AI and use AI to meaningfully operate their business. When it’s real, the output shows it: small teams shipping at a pace that would have required 10x the headcount five years ago.

But most companies who claim to be here aren’t. For a surprising number of them, it’s more aspirational than representative. More on this in the next section.

Fake AI (AI, Non-AI)

These are companies that sell a product built on AI but are NOT using AI to meaningfully operate their business.

These companies are problematic for a couple of reasons.

1. Incoherence – They believe in AI enough that it’s the competitive advantage of their product but apparently not enough to use it meaningfully themselves? It’s like a priest who doesn’t practice.
2. Cursory Understanding – AI is moving so fast if you don’t use it daily, in a meaningful way, you won’t understand it well enough to build a company around it.
3. Bad Signal – AI is widely available to almost anyone. If you’re an AI company (or any company) that doesn’t use AI, it’s because you’re choosing not to. As a founder, you need every competitive advantage you can get. It’s a bad sign if you’re turning one down.

For most of these companies, nothing nefarious is going on. Often, the founders don’t know enough about what’s possible to realize how they use AI is cursory at best. When we talk to these founders, it’s easy to tell. If they talk about using ChatGPT, we know they aren’t that sophisticated. If they can’t talk in detail about their agent harness, how they’re managing shared context, etc. we know they can’t be doing meaningful work with it.

If deep down, you think your company might be in this quadrant, start building things with AI, even if it’s just for fun. The best way to understand the technology is to use it.

Traditional (Non-AI, Non-AI)

These are companies that do NOT sell a product built on AI and are NOT using AI to meaningfully operate their business.

Not much to say here except that the vast majority of businesses exist in this quadrant. Wherever you are right now, look at the businesses around you. They are probably all traditional businesses (restaurants, banks, etc.). If you’re in tech, or read what companies get covered in the news, you mainly hear about tech companies and forget that almost all businesses are traditional businesses.

Stealth AI (Non-AI, AI)

These are companies that do NOT sell AI but are using AI to meaningfully operate their business.

True AI companies get most of the attention and rightfully so, but this quadrant has a lot of quiet potential. The ceiling on the value of these companies is an order of magnitude smaller than the ceiling of True AI companies (I don’t think any of these companies can be worth the $1T Anthropic and OpenAI are being valued at) but my hunch is there’s room for a lot more of them.

There are two ways to start these types of companies: (1) start the company like you would any other company and (2) start by buying an existing business that fits the target. The former is standard. The latter is interesting.

Buying businesses and using technology to make them run more efficiently isn’t new. PE firms like Vista Equity Partners built $100B+ businesses doing software roll-ups: buying enterprise software companies, standardizing operations, and expanding margins. It works, but only when (1) the work was structured enough (data, workflow, etc.) for software to handle and (2) the opportunity was lucrative enough to build software for.

AI removes both constraints. AI handles unstructured work, and it’s made software so cheap to build, you can apply it to a wider range of opportunity sizes.

A few recent examples.

1. Crosby – AI-first law firm backed by Sequoia and Bain, with a median contract review time of 58 minutes. They built from zero with AI as the default operating model.
2. Long Lake – Backed by General Catalyst, is rolling up HOA management companies: 18 acquisitions, $1.28B raised, $100M in EBITDA in under two years. They bought businesses with proven revenue and used AI to transform margins.
3. Crete Professionals Alliance – Backed by Thrive Capital and Bessemer, is rolling up U.S. accounting firms: 20+ acquisitions, $300M+ in annual revenue, with OpenAI-powered tools saving audit teams hundreds of hours per month.

Customers of these companies have no idea AI is involved. They shouldn’t. What they notice is faster turnaround times, better service, and competitive rates.

The obvious counter-argument is if the moat is “we use AI to run operations better,” what happens when every competitor has access to the same AI tools?

A few reasons why I’m not that worried about it.

1. Slow Adoption – Traditional companies are slow to adopt even when the competitive advantage is obvious. Think about how many companies still aren’t fully using software (especially in Southeast Asia) 20 years after it became widely available. The tools aren’t the bottleneck. The organizational capacity to use them is. More on this in the next section.
2. Effort to Implement – Making AI work in a specific industry isn’t plug-and-play. It requires deep domain context, refined workflows, and operational knowledge that compounds. The moat isn’t “we have AI.” The moat is “we’ve spent years making AI work in this specific business.”
3. Data Compounds – The longer you operate in a specific domain with AI, the more proprietary data you generate: edge cases, customer patterns, performance benchmarks. That data feeds back into the AI and makes it better. A new entrant can’t shortcut years of accumulated data.

Why Traditional Companies Stay Put

Why don’t Traditional companies just adopt AI themselves? The tools are available. The cost savings are clear. Four structural forces make internal transformation nearly impossible for most of these companies.

1. Risk Asymmetry – Nobody gets fired for doing things the way they’ve been done. People get fired for trying something new and having it fail. The rational career move is to wait.
2. Incentive Mismatch – The payoff of margin expansion accrues to the company. The risk, time, and political capital accrue to the person who pushed for it. The math doesn’t work for the individual.
3. Authority Gap – Most people don’t have the authority to push change of this magnitude, including most people near the top. The number of people who both understand AI and have the authority to implement it at scale is usually zero.
4. Security – Data, IP, and regulatory compliance are legitimate concerns. They’re also the most convenient veto for people whose real objection is one of the first three.

These four forces don’t just slow AI adoption. They explain why software didn’t fully eat these industries either. The transformation, when it comes, mostly comes from outside.

Migration

The quadrants aren’t static. Companies move, and in the cases of Traditional and Fake AI, they should. The interesting paths are all about getting out of those two quadrants.

1. Traditional → Stealth AI – The most valuable move in the framework. This happens internally if a traditional company has strong leadership and absolute belief in making the leap. If not, that’s where the opportunity for acquisitions and roll-ups is. New ownership brings the authority, incentive alignment, and risk tolerance to change. It’s also the path for greenfield builders who start an AI-native business in a Traditional industry from scratch.
2. Traditional → Fake AI – The temptation path. Bolt an AI label onto the marketing without changing operations. Easiest move to make, worst place to end up. You attract AI-level competitive scrutiny without the operational advantage to survive it.
3. Fake AI → True AI – This is easier than it sounds, which is what makes it frustrating that more companies don’t do it. The tools are available. The playbooks exist. The barrier isn’t technical. It’s the same inertia and poor judgment that put you in the Fake AI quadrant in the first place.
4. Fake AI → Stealth AI – If you’re honest about being in the Fake AI quadrant, this might be the better path. Drop the AI product and apply your operational AI knowledge to a traditional industry. The bar for Stealth AI is competence with AI, not breakthroughs. Most of the time, you don’t need to be True AI, and the Stealth AI opportunity is much larger.

The Southeast Asia Opportunity

SEA has a lot of Traditional companies for two reasons.

1. Regional Fragmentation – Ten countries, dozens of languages, radically different regulatory and operating environments.
2. Lower Technological Sophistication – Most SEA businesses are small, owner-operated, and never had the scale pressure, capital, or vendor ecosystem to digitize. Many skipped the software era entirely.

These typically look like impediments to building startups in Southeast Asia, but for Stealth AI opportunities, they’re advantages. Think about the companies that do things like property management, logistics, freight forwarding, accounting, BPO, customer support, F&B operations, healthcare administration.

These are the industries where the work is labor-intensive, the businesses are fragmented, and AI-enabled consolidation has the most room to run. The fragmentation that usually looks like a problem is the setup: the same AI-enabled operational playbook that consolidates a property management business in Jakarta works in Kuala Lumpur, Manila, and Ho Chi Minh City, just with different execution.

For founders in SEA, the question is which Traditional industries to move into the Stealth AI quadrant before someone else does.