Durable Software vs Dead Software Walking

Everyone's asking the wrong question about AI and software.

The question isn't "Will LLMs disrupt software?" They already are. The question is: which software survives?

After analyzing 21 companies through the lens of structural moats — not brand, not TAM, not vibes — a clear pattern emerges. Software durability in the age of AI comes down to three binary questions. Get all three right, and LLMs become your distribution layer. Get all three wrong, and you're a wrapper waiting to be unwrapped.

The Three Hard Moats of Software

Here's the framework. Three questions. Yes or no. No hand-waving allowed.

1. Is the data proprietary? If yes, the moat holds. If no, the accessibility layer is collapsing. LLMs are the greatest data-democratization engine ever built. If your software's value proposition was "we organize publicly available information into a nice interface," congratulations — you just described ChatGPT's weekend project. But if your data can't be replicated because it was generated inside your ecosystem — credit scores derived from decades of lending outcomes, healthcare records locked behind compliance walls, social graphs built by billions of users — then LLMs can't replace what they can't access.

2. Is there regulatory lock-in? If yes, LLMs don't change the switching cost equation. If no, switching costs are primarily interface-driven and dissolving. Regulation is the moat that AI can't code around. When the government says "you must use an NRSRO-rated agency" or "patient records must flow through a HIPAA-certified EHR," no amount of prompt engineering changes that reality. Regulatory lock-in is the only moat that doesn't require continuous reinvestment — the government does the maintenance for you.

3. Is the software embedded in the transaction? If yes, LLMs sit on top of you, not instead of you. If no, you're replaceable. This is the subtlest and most important distinction. If your software is the pipe through which money, trades, or critical workflow actions flow, AI becomes your copilot, not your competitor. Stripe doesn't fear ChatGPT because ChatGPT still needs Stripe to move money. But if your software is the lens through which people view data — a dashboard, a search tool, a reporting layer — then you're one good AI interface away from irrelevance.

The scoring is brutal in its simplicity:

Zero "yes" answers → No moat. You're a feature, not a product. One "yes" answer → Weak moat. Survivable, but vulnerable. Two or three "yes" answers → Strong moat. LLMs are additive, not existential.

The Durability Table: 21 Companies Ranked

What the Table Reveals

The Fortress Four — FICO, Moody's, Epic, and Veeva — represent the gold standard: proprietary data generated through decades of industry-specific activity, regulatory frameworks that mandate their use, and direct embedding in the transactions their customers execute. FICO scores don't just inform lending decisions — they are the lending decision. Moody's ratings aren't advisory — they're required for bond issuance. No LLM changes this equation.

The "Strong" tier is where it gets interesting. Meta has no regulatory lock-in (regulators are arguably working against it), yet its social graph is the most proprietary dataset on earth and its advertising platform is the transaction itself. Interactive Brokers and Robinhood benefit from broker-dealer licensing requirements and direct trade execution embedding — ChatGPT can analyze a stock, but it still needs a regulated broker to buy it. Roblox's entire UGC ecosystem and virtual economy constitute a proprietary world that AI can enhance but not replicate.

The "Moderate" tier contains the companies most investors assume are safe but shouldn't sleep easy. Bloomberg's terminal data is proprietary, but it's not embedded in the transaction — it's the lens, not the pipe. Salesforce's CRM data creates stickiness, but that data is increasingly portable and the interface layer is exactly what AI disrupts best. Palantir sits in a fascinating middle ground: government security clearances create partial regulatory lock-in, and its operational embedding is deepening, but it owns none of the underlying data.

The "Weak" and "HIGH RISK" tiers are where the AI disruption narrative is most real. Netflix and Reddit have genuinely proprietary data (original content and UGC respectively), but neither is embedded in transactions nor enjoys regulatory protection — they're pure interface plays. ServiceNow's workflow embedding provides some protection, but without proprietary data or regulatory moats, a sufficiently capable AI orchestration layer could replicate its core value. And at the bottom — FactSet, HubSpot, Spotify, LexisNexis search — these are essentially accessibility layers sitting on top of non-proprietary data with no regulatory protection and no transaction embedding. This is exactly the category LLMs were built to disrupt.

The Takeaway

The companies at the top of this table share a common trait that goes beyond software quality: they are infrastructure, not interface. The distinction matters because AI is the greatest interface improvement in computing history. If your software's value was primarily in how it organized and presented information, AI just commoditized your entire business model. But if your software is the regulatory-compliant, transaction-processing, proprietary-data-generating backbone of an industry, then AI just became your most powerful feature.

When evaluating any software position in your portfolio, run the three-question test. If you can't answer "yes" to at least two, you're not owning a compounder — you're renting an interface.

Would you like to stay ahead of opportunities like this? Join our community where we share real-time trade alerts and deep-dive analyses of businesses with true competitive advantages. Don't just trade the market - invest in excellence.

Want to receive our trade alerts and detailed analysis in real-time? Join our community of value investors who understand that pricing power is the ultimate competitive advantage. Receive our trade alerts on your phone? Download the app here