The Meat Moat

Can an 89-year-old economic theory solve the AI software conundrum?

The Meat Moat

Software is getting cheaper and cheaper to build. It is reasonable to expect that, within the next 2 years, an AI system such as Codex, Cursor, or Claude will be able to autonomously create high-quality, medium-sized software products.

But humans are not going to get any faster! They will still take time to learn new skills and processes, make purchasing decisions, trust new vendors, change workflows, and so forth. Similarly, humans will still be humans—social creatures at heart! The social ties that bind us together will still exist, be it family, friends, or organizing and collaborating on any number of topics.

Just as the cloud created a whole new set of winners and losers in the software industry, AI will create a new set of winners and losers. The winners will be building “meat moats”—businesses that are durable because humans are slow, because we are social, and because trust, liability, adoption, and coordination do not move at the speed of code.

When code becomes abundant, the scarce resource is not software. The scarce resource is human behavior and how that manifests in trust, adoption, compliance, accountability, distribution, collaboration, and habit.

SaaSpocalypse

In the past few months, there has been a lot of discussion online about the “SaaSpocalypse” and the contrarian view of “Software will eat AI” from HSBC and others.

Much of this has been fueled by the dramatic improvements in AI tools’ ability to write software. What happens if software can be quickly written by AI? The SaaSpocalypse argument holds that software firms, or at least many of them, will go away as it becomes trivially easy to write software. The contrarian view argues that enterprise software needs to be extremely reliable and predictable, and that LLM-based AI is inherently non-deterministic and ill-suited for the task.

Both views are right in a sense, or, perhaps better put, both make valid points. That does not help, though, if you are trying to make investment decisions!

A better lens to analyze this is a seminal economic theory from 1937: Ronald Coase’s The Nature of the Firm. Why do companies even exist? Why not just have people, with everything being transactions between people? Or, as applied to SaaS companies, why do we even need software companies if it will soon be possible to vibe code any app you desire?

The answer is transaction costs.

Coase’s theory of the firm holds that firms exist when organizing work within a firm is cheaper than coordinating everything through market transactions. It is not that people are incapable of doing each individual task themselves. It is that the overhead of search, coordination, trust, quality control, accountability, and decision-making becomes too high.

Let’s start with restaurants as a business. Sure, I can grow my own food, harvest it, and cook it. I could even hire different people each day to do these tasks in my home garden and kitchen. But it sure is easier to go to a restaurant to eat!

Restaurants exist because, even though I technically have the ability to grow, harvest, and cook my own food, it is much more convenient to go to a restaurant. Importantly, a restaurant is more than just food; it offers an experience, be it the ambiance, decor, location, service, etc. Even if I had a home robot that could cook any dish I desired, I would probably still go out to a restaurant for date night!

A restaurant encapsulates hundreds of behind-the-scenes decisions and actions (location, menu, ingredient shopping, cooking, etc.) into one simple “pay money, get food and experience” transaction.

As a human, that simplicity is valuable. A restaurant makes my decision-making easier. It creates a space for social interactions. Those things will continue to be valuable, even in the coming age of robots.

Coase’s theory explains why that kind of packaging matters: a firm (such as a restaurant) can reduce the transaction costs a customer would otherwise have to pay in time, attention, risk, coordination, and trust.

The Nature of Software Companies

If we apply this theory to software companies, we can easily identify numerous areas where transaction costs were minimized historically:

• Writing software was hard and took time.

• Software could only be written by trained engineers; finding them was hard and expensive.

• Software requires continuous investment to maintain functionality and address defects.

• Selling and supporting software was hard.

The first three are increasingly no longer true—it is getting easier and easier to write software. That is the argument for the SaaSpocalypse. But the humans who use software are still there!

That’s the meat moat. Anyplace where humans are interacting with software will create new places of friction, the new boundaries that will constitute a defensible and durable software company.

These include:

• Regulations and compliance. One could argue that government regulations will be ASI-proof (artificial super-intelligence). As long as there are humans, there will be governments. As long as there are governments, there will be bureaucracies! And as long as there are bureaucracies, there will be regulations.

  For example, if you want to sell software to the government, you need to get CMMC certification. Cyber AB marketplace tracks the authorized CMMC Third-Party Assessor Organizations (C3PAOs) required for many defense-industry cybersecurity assessments. There are relatively few assessment organizations, and becoming CMMC-certified is time-consuming and difficult. It does not matter how cheap the software is to build if the business value depends on a regulatory gate, an approved assessor, and a human institution accepting the result.
  

• Platform effects. Let’s assume an absolutely perfect future version of Codex that can build any software an individual or company might want. Great! OK, now what happens when the real value comes from hundreds, thousands, or even millions of other people needing to use that software?

  As a simple example, consider something like Windows as an operating system. It is valuable because nearly every piece of hardware works with it, and millions of applications run on it. AI coding tools will eventually be able to build an OS, but who will make sure the OS works with all the different printers, computer hardware, networking systems, etc.? Who is going to train hundreds of millions of users on a new OS?

  Again, it does not matter how easy it is to build the software on its own if the value comes from millions of people using it.
  

• Adoption and change management. AI can write software quickly. It cannot instantly make a company change how it works.

  Anyone who has sold or deployed enterprise software knows this. The hard part is often not the code; it is getting budget approval, passing security review, integrating with existing systems, training employees, changing incentives, updating processes, and getting people to trust the new workflow. Even when better software exists, companies keep using the old system because the organizational switching cost is real.

  That is a meat moat. The software vendor that helps customers cross the human adoption gap can be far more valuable than the vendor that merely ships features.
  

• Distribution and customer intimacy. There are literally tens of thousands of enterprise software companies, covering everything from marketing lead generation to managing the machinery on a factory floor, to order processing.

  Even in a space as “narrow” as order processing, there are literally dozens of subcategories: invoice processing software, accounts payable processing software, software for quotes and configuration—I could go on! A typical large enterprise might use hundreds of these specialty applications—thousands in some cases. In one large global company I’m familiar with, they literally had over 5,000 distinct software tools used to run their back office operations.

  This fragmentation makes sense in a world where software is expensive to build. Solving any particular piece of the puzzle was hard, and often only a handful of companies would go after it.

  In the AI era, a company that is both a) incredibly fast at writing software features and b) incredibly adept and efficient at understanding, reaching, selling to, and serving customers is going to be dominant. Anytime a customer asks, “Can you do X?” the answer will be “yes,” and that capability shows up literally the next day.

  These companies will not be selling software so much as “problem-solving as a service.”
  Up to now, IT and other employees at a company have done the problem-solving, and then would essentially buy the hammers and nails (B2B software tools) needed to fix the problem.

  Going forward, the most capable software vendors are going to move upstream and solve problems for customers, whether it’s taking care of the books, processing orders, finding customers, etc.
  One of my favorite AI tools for small businesses these days is Puzzle. It is an AI-based accounting tool that takes care of the books for a small business without you really needing to think about it. It connects to my bank, Stripe, etc., and, very literally, manages all the bookkeeping with AI automation. It’s replaced both my bookkeeper and QuickBooks (in QuickBooks, I still had to do the work! It’s not automatic).
  

• Liability moats. One of the strongest meat moats will be those involving liability and expertise.

  This year, I ran my personal tax preparation through Grok. To set some context, my personal federal tax returns usually run about 100 pages of filings, so needless to say, there are even more pages of source material that feed into those filings. Certainly not the most complicated out there, but definitely not a quick job either.

  Grok did a phenomenally good job. It literally came within 0.2% of the same number as my professional tax accountant—and it did it within minutes after I gave it all of the source materials.

  It is easy to see that future versions of Grok would be even better.

  Given that trajectory, would I ever want to save money and do my own taxes again?

  Probably not.

  It is not a question of whether there will be tools good enough to help. There definitely will be. It is a question of liability.

  There is so much complexity in those 100 pages of filings. I’m certainly smart enough (or at least I would like to think that I’m smart enough!) to figure it out, given enough time and effort. But I don’t want the liability of being wrong. I’d much rather hire the “expert” (in this case, my certified tax accountant) and rely on their expertise. If there ever was a problem, they would be the ones to take care of it.

  The U.S. tax system is sprawling: Title 26 alone runs thousands of pages in GovInfo, and that is before you add Treasury regulations, IRS guidance, forms, cases, and professional commentary. Going back to Coase’s theory of the firm, it is easier and safer for me to rely on a third party to handle taxes and liabilities.

  There are many aspects of life and business where it makes sense to transfer liability or accountability to another company: legal matters, financial matters, and healthcare. It may be possible in the future for a robot to perform open-heart surgery, but I think even in that future world, I would still want a hospital to perform the surgery rather than do it myself!

The vulnerable software companies are those whose only moat is that the software was hard to build. The durable ones are the companies where the software is only part of the value: regulated workflows, trusted brands, complex adoption paths, ecosystem participation, human accountability, and distribution. AI compresses engineering time. It does not compress human trust nearly as quickly.

There will undoubtedly be many more meat moat categories, and certainly new businesses and new moats we can’t even dream of yet. SaaS companies that don’t evolve will likely struggle or die. But as I’ve outlined above, even in the world of AI, Coase’s theory of the firm still applies. Companies will continue to exist to minimize transaction costs, and some number of those companies will use software to solve problems and deliver results to their customers.

The old SaaS model was: sell software that helps customers solve problems.

The new SaaS model may be: use software, AI, service, trust, and accountability to solve the problem outright.

The code moat is shrinking. The meat moat is just getting started.