TL;DR: Every founder I meet is paralyzed by the same anxiety: "Is my product just a ChatGPT wrapper?" or "Won't Google just kill me next week?" These fears are valid—even the smartest kids at top universities are hesitating to build because of the looming shadow of incumbents. But the concept of the "Moat" hasn't died; it just evolved. If you stop looking at the code and start looking at the Business Architecture, there are seven counter-intuitive ways to survive.
James here, CEO of Mercury Technology Solutions. Taipei - December 30, 2025
The narrative that "there are no moats in AI" is lazy. Based on what we are seeing from Y Combinator and the frontlines of Silicon Valley, the defensive barriers are shifting from Technology to Operations and Economics.
Here is the blueprint for building an indefensible business when the core tech is a commodity.
1. Speed is the Only Moat That Matters Early On
Forget patents. Forget 5-year plans. In the early stage, your only advantage is that you are not Google. Cursor (the AI code editor) is the perfect case study. They operate on a "daily sprint" rhythm. They reset the clock every 24 hours to ship new features. Google and OpenAI cannot do this. Before they ship a button, it has to go through Product Managers, Legal, and PR reviews. As Paul Graham says: "In the early stage, speed is the only moat."
2. Build the Castle, Then Dig the Ditch
Founders are obsessed with defense before they have anything to defend. The YC "Golden Rule" is simple:
- 0 → 1: Build a Castle (Solve a painful problem).
- 1 → N: Dig the Moat (Build defense).
If you haven't built a solution that solves an "existential pain" for a customer, your moat is just a puddle in an empty field. Stop optimizing for protection and start optimizing for value.
3. Process Power: The "Schlep" Advantage
Hamilton Helmer’s concept of "Process Power" is back. A hackathon demo can get you to 80% accuracy with minimal effort. But enterprise clients need 99% accuracy, and bridging that gap requires huge, boring engineering effort.
- Greenlight (KYC) & Casa (Loans): They can't afford hallucinations because a single error costs millions.
- The Moat: Many engineers suffer from "Schlep Blindness"—they hate the tedious work of edge-case handling. If you are willing to do the dirty work to get to 99%, you win.
4. Switching Costs 2.0: Deep Integration
Data migration used to be the lock-in. Now, AI makes moving data easy. The new lock-in is Workflow Integration. Look at Happy Robot and DHL. They spent 6 to 12 months wiring AI agents into DHL’s complex logistics web. Once that system is live and generating value, DHL isn't going to rip it out to "audition" a competitor. For consumers, the moat is Memory. If an AI knows your context and history, leaving it feels like getting a lobotomy.
5. Counter-Positioning: Attack the Business Model
This is how you kill a giant: Do something they cannot do without destroying their own revenue. Legacy SaaS loves "Per-Seat Pricing." It’s their addiction.
- The Conflict: If an AI agent works, you need fewer humans. Fewer humans means fewer seats. Salesforce hates this.
- The Attack: Startups like Aoka (HVAC industry) charge for "Work Done" (Outcome-based pricing), taking a 4-10% cut of the transaction. They align with efficiency; incumbents align with bloat.
6. Network Effects: The Data Flywheel
"User Scale" is the old metric. The new metric is the Data Flywheel. The loop is: Usage → Data → Optimization → Better Product.
- Cursor: Records every keystroke (even on the free plan) to train its completion model.
- Salient: Partners with banks to get private evaluation data (evals) that no open model has access to. You aren't building a user base; you are building a proprietary dataset.
7. Economies of Scale (App Layer)
We think Scale Economies only apply to training massive models. Wrong. You can build scale at the Application Layer. Exa.ai built a massive static index of the web specifically for AI search. This required huge upfront capital (CapEx). But now that it's built, they can serve thousands of clients at a near-zero marginal cost, while a new competitor faces a massive wall of initial investment.
Conclusion: Look Beyond the Code
The era of "Tech-First" moats is ending. The era of "System-First" moats is beginning. Whether it’s relentless speed, boring reliability, or pricing model disruption, the opportunities to build an empire are there. You just have to stop staring at the LLM and start staring at the business.
Mercury Technology Solutions: Accelerate Digitality.
