Building Moats in the Age of AI: Insights from StartX VC Partners

 
Primary focus of the round table: AI’s rapid evolution raises the question of durability for every new AI-native Startup. Are traditional moats the answer? Or is there a new type of moat?

Introduction: Stanford’s StartX recently hosted a panel discussion titled “What is Investable in AI?”, featuring venture investors from Andreessen Horowitz, Lightspeed, and Altos Ventures. The conversation (led by StartX’s Head of AI, Andrew Radin) centered on how AI startups can build durable businesses in a landscape where technological advantages are often short-lived. All the panelists agreed that in the era of generative AI, building a “moat” – a sustainable competitive advantage – is harder and more crucial than ever.

Traditional moats have historically protected businesses, but a new concept introduced by Lightspeed’s Guru Chahal – the “system of engagement” – is a new era moat.

Traditional Moats and the Durability Challenge

The Race and the Risk: In the fast-moving AI arena, product durability is a chief concern. It’s easier than ever for competitors to replicate features or models – sometimes in a matter of weeks – thanks to open-source models and readily available APIs.

“When your core technology is available to anyone with an API key and a credit card, traditional technical advantages may look minimal.”

In other words, simply having a better model or algorithm is rarely enough; today’s breakthrough can become tomorrow’s commodity. This dynamic has led some to lament that it feels like “there are no more moats to build” in AI. Yet, the panelists emphasized that moats do still exist – they’re just shifting from pure tech toward business and user-oriented realms.

Classic Moats Revisited: The investors first pointed to well-known strategy frameworks (like Hamilton Helmer’s “7 Powers”) which enumerate classic moats that companies can cultivate. A few of the traditional moats discussed include:

• Network Effects: When each additional user adds value to the product for all users, it creates a self-reinforcing advantage. This is a durability factor in many tech companies – for example, a social platform or marketplace becomes more useful as its user base grows, making it hard for a new entrant to catch up. In an AI context, one might think of community-driven platforms (for instance, AI that relies on user-generated content or feedback) where a growing user base improves the service for everyone. A pertinent example is CharacterAI’s ecosystem of user-created AI personas – with over 16 million characters made by users, the content library itself attracts more users, forming a network effect moat.
• High Switching Costs (Systems of Record): If a product becomes deeply embedded in a customer’s workflow or data infrastructure, switching away from it is painful – this lock-in is a classic moat. In enterprise software, being the “system of record” for critical data has long been a durable advantage. “Previously, the focus was on Systems of Record – think ERP, CRM, HRMS. Companies found their competitive moat in owning these monolithic systems that stored critical data.” Once a hospital, for instance, has all its oncology records in Flatiron’s system (as panelist Jay Rughani’s former company did), or a sales team has years of customer info in Salesforce, it’s extremely hard to rip and replace. This moat comes from workflow integration: the product becomes so ingrained in daily operations that alternatives face a steep uphill battle to dislodge it.
• Scale Economies: Economies of scale – doing things at a volume that newcomers can’t match – can protect a business by making it the lowest-cost or most efficient player. In AI, scale can manifest in data and compute: a company that has amassed a unique, proprietary dataset or the means to train models at a massive scale might enjoy an edge that others can’t easily copy. For example, a large language model that has been fine-tuned on years of user interaction data from a particular domain has a knowledge and performance advantage (data network effects) that a fresh competitor won’t have on day one. The panel noted, however, that these advantages can erode quickly as technology diffuses and open-source projects narrow the gap.
• Brand Trust and Reputation: A strong brand can act as a moat by itself – especially in an era where AI tools occasionally hallucinate or make mistakes. Users will gravitate towards providers they trust. The panel cited that trust is essential in AI; companies that establish a reputation for reliability and accuracy can build loyalty that newcomers struggle to undermine. (Think of how OpenAI’s ChatGPT became a household name for AI chat – that brand recognition and trust gives it a defensive buffer, even as alternative chatbots appear.)

In fact, the panelists stressed that smart founders should try to combine multiple moats if possible. For example, an AI healthcare startup might aim to become the system-of-record for patient data and leverage network effects via a physician user community, all while building a trusted brand for accuracy and compliance. The challenge, however, is that in the current wave of AI, many of these advantages are harder to defend. Powerful open (and closed) source models, advanced agentic capabilities speeding up development time, and all of these tools being increasingly democratized means a determined competitor (or incumbent) can often replicate your technical features quickly – and if they have a strong cornered resource or just the right positioning, there could be a black swan event for startups that have not found the correct strategic positioning.

Introducing the “System of Engagement”

In addition to these “old school” moats, Lightspeed partner Guru Chahal introduced a new moat: the “System of Engagement.” This term flips the focus from owning data or workflows (systems of record) to owning the relationship with the user. While the phrase “system of engagement” (SoE) has appeared before in enterprise strategy (often referring to user-facing interfaces on top of back-end systems), Guru’s usage reframes it as a durability moat for AI products specifically with an equal focus on B2B.

So what is a system of engagement in the AI context? It’s essentially about creating a deep, personal interaction layer between the AI product and the user – a feedback loop that gets stronger over time. A system of engagement means the product isn’t just a tool that a user can swap out; instead, it becomes an adaptive companion that users grow accustomed to and dependent on. The AI continuously learns from the user’s behavior and data, tailoring itself to their needs and preferences, while the user in turn becomes increasingly comfortable with the AI’s unique quirks, interface, and “personality.” The result is a two-way bond that is hard for a competitor to break.

To make this concrete, Guru gave examples of how this plays out in both consumer and business settings. In the B2C realm, consider how quickly millions of people integrated ChatGPT into their daily routines for writing, coding help, or brainstorming. Part of ChatGPT’s stickiness is simply first-mover advantage and brand, but part is also the ongoing engagement: users have lengthy chat histories, custom instructions, and familiarity with how ChatGPT responds. Over time, many users have effectively “trained” their ChatGPT (through those ongoing conversations and feedback) to suit their style or recall their prior context – and more importantly they have been trained on ChatGPT's shortcomings and they have learned to adapt to these.

In short, the feedback loop here is a two sided coin – not only is ChatGPT learning about each user but each user is learning ChatGPT – this creates a very sticky feedback loop.

Why Engagement Builds Moats

Taking a step back, moats in AI are notoriously difficult:

• Models commoditize quickly — the same LLM can be accessed via APIs.
• Features can be copied — summarization, transcription, chatbots.

But an SoE moat comes from:

1. High switching costs: If your platform becomes where workflows live, users won’t abandon it easily (data, habits, integrations).
2. Continuous data advantage: Engagement surfaces proprietary usage data (corrections, preferences, edge cases) that improve your models uniquely.
3. Embedded trust & brand: End users associate your tool with reliability, not the underlying model provider.
4. The user learning the brand: This is not just switching costs where users setup workflows; this is the other side of the feedback loop. A user has actually learned to intuitively assess the non-deterministic quirks of an AI tool. This may be the stickiest of all moats.

Durability in AI via SoE

Durability comes from resilience to model commoditization:

• If your moat is just 'we fine-tuned an LLM', that’s fragile.
• If your moat is 'we own the workflow where the LLM is embedded', you control the feedback loops, context, and adoption. That’s durable.
• If your moat is 'I have such a high touch point with the user that they adapted their workflow to better navigate my version of AI'; that may be the stickiest moat.

A New Type of Lock In

Switching to a new AI assistant would mean losing those conversation histories and starting from scratch; means learning the a new set of shortcomings for a new model – a subtle but real switching cost. Likewise, developers who’ve gravitated to Anthropic’s Claude for coding might cite its 100K-token long context window and helpful code explanations as something they now rely on. Once you’ve experienced an AI that can ingest your entire codebase and deliver answers in one go, other models that can’t do the same feel inferior; you’ve engaged deeply with Claude, and your workflows now assume its capabilities.

These are examples of engagement-driven lock-in: it’s not that the underlying model can’t be reproduced by others, but the user-product relationship gives it an edge.

In fact, engagement moats have already emerged in consumer AI. An illustrative case is Character.AI, a chatbot platform where users create and chat with AI characters. Character.AI focused on fun and emotionally engaging interactions (entertainment and role-play) rather than generic Q&A – and it paid off in fierce user loyalty. Teens and Gen-Z users spend hours on it, to the point that they dismiss other chatbots. As one observer noted, when asked if they’d tried ChatGPT, a group of teen users scoffed: “Why would we use that? It’s not fun.” They preferred Character.AI because of its personality and experience – essentially, they became attached to the way that specific AI engages them. This kind of emotional or habitual engagement can form a moat: even if another model has equal technical ability, the users don’t want to leave the experience they love.

Perhaps more surprisingly, enterprise (B2B) software can also harness systems of engagement as a moat – and Guru Chahal emphasized this is happening already. In business settings, a “system of engagement” might involve an AI tool becoming woven into every step of a company’s operations, from onboarding new users, to daily use, to ongoing customer success touchpoints. For example, take the healthcare AI startup Abridge: Abridge provides an AI “medical scribe” that listens to doctor-patient conversations and automatically generates clinical notes. It’s a sophisticated technology, but importantly, it directly engages the physician during their routine patient visits – essentially becoming an ever-present assistant in the exam room. Doctors using Abridge quickly become accustomed to the AI’s presence and the way it formats their notes. The AI in turn learns the doctor’s specific vocabulary and preferences (customizing how it summarizes based on the doctor’s style over time). If a hospital tried to swap in a different note-taking AI, they’d face resistance not only due to customization migration, but because the physicians have grown comfortable with Abridge’s workflow integration and output style – in essence they have learned to handle the quirks of Abridge. In this way, user engagement becomes a moat: the product is tuned to the user, and the user is in tune with the product.

Another B2B example is the success of tools like Slack in unseating entrenched incumbents. Slack didn’t win enterprise chat by having wildly unique technology – it won through superior engagement and UX. It “reinvented how chat felt” – with a responsive, delightful interface that users actually enjoyed – and made itself indispensable to their daily work life. Users adopted it enthusiastically (often without top-down mandates), and it became habit-forming. Slack’s design led to “deeper emotional engagement” and a product people would fight to keep using. In effect, Slack became the system of engagement for workplace communication, displacing earlier tools despite those incumbents having the data and records (systems of record). Guru’s point was that AI startups can aim for a similar outcome: even if your product plugs into an existing system of record, own the engagement layer with the user through superior experience and personalization. If your AI is where the user spends their time and derives joy or efficiency, that position can be as defensible as any hard tech moat.

What makes the system-of-engagement moat especially relevant now is that technical differentiation in AI is often transient, but a strong user bond is harder to steal. A rival can copy your model’s architecture or use the same open-source backbone, but they cannot clone your community’s loyalty or your users’ habits. As Greylock’s Jerry Chen wrote, “Systems of engagement are the interfaces between users and the systems of record, and can be powerful businesses because they control the end user interactions.” Users might try other options, but if they’ve grown to love the workflow, recommendations, and even idiosyncrasies of your AI product, they’ll be reluctant to switch. Moreover, controlling the engagement layer gives a company valuable data and feedback. Each interaction with users generates data that can improve the product further – preferences, behavior patterns, fine-tuning signals – which in turn makes the user even more attached to the increasingly personalized experience. This creates a virtuous cycle: engagement leads to a better product, which leads to deeper engagement. Over time, as Chen notes, a startup that owns the system of engagement can even parlay that into becoming a system of record too, by virtue of all the data it accumulates. It’s a long-term strategic path: start by owning the user interface and relationship, end by also owning a trove of proprietary data.

Why the “System of Engagement” Moat Matters

The introduction of engagement moats isn’t just semantics – it signals a mindset shift for founders and investors in the AI space. Traditionally, when evaluating a startup’s defensibility, investors asked questions like “Do you have unique data or superior algorithms?” Those are still important, but this panel made it clear that how users interact with the AI is just as critical in determining long-term success. Here’s why this concept matters:

• Harder to Copy than Features: In a world where features can be copied or new models leapfrog old ones within months, focusing on user engagement targets something a fast-follower can’t replicate overnight. A competitor might spin up a similar AI service, but they won’t have your users – especially if you’ve spent time nurturing those users, adapting to their needs, and maybe even building a community around your product. That human connection and habit formation can act like fortress walls. It's said, “In a world of software abundance, experience is the only scarcity left.” If your AI delivers a vastly superior experience (e.g. more intuitive, more enjoyable, more tailored), users will not readily go to a bland competitor. Good design and engagement build habit loops and turn users into champions of your product, creating emotional switching costs that augment any technical switching costs.
• Competitive Advantage in Adoption and Retention: The system of engagement concept also offers a playbook for gaining market share. If your AI product is highly engaging, it likely means it’s easier to adopt and stick with. Users onboard quickly because the product draws them in (as opposed to requiring heavy training or being a chore to use). Engaged users tend to talk about the product to others, fueling organic growth. And they tend to stay. High retention is a natural outcome of an engagement moat – when users feel the product is almost a partner or an extension of themselves, churning out becomes unattractive. We see this in metrics: products that emphasize intuitive UX and user-centric design boast faster user adoption and lower churn than products that compete purely on technical specs. For AI startups, this implies that investing in UX, personalization, and customer success can pay greater dividends for defensibility than, say, a slight model accuracy improvement that users might not even notice.
• Applies Across the Customer Journey: One powerful aspect of the engagement moat is that it can be reinforced at every stage of the customer journey. The panel discussed that this isn’t just a front-end UI issue; it spans from onboarding (make the first user experience delightful and tailored) to customer support and success (continue engaging the user with helpful touch points, updates, and improvements). In fact, even outside of AI, SaaS experts consider strong customer success a “human moat” – because competitors can copy your software, but not the personal relationships and knowledge you build with customers. An AI product that, for example, onboards new users with an interactive tutorial custom-tailored to their goals, and later provides proactive support (perhaps an AI agent that checks in on your progress or auto-adjusts settings to your usage), is creating engagement at each step. This comprehensive approach makes the user feel “looked after” and further cements loyalty. Guru Chahal highlighted that we’re seeing AI companies execute on this in B2B: from the moment a customer signs up, through their daily use and any support needs, the successful startups keep the user continuously engaged and learning, rather than treating the product as a one-and-done tool.

In summary, the “system of engagement” moat matters because it highlights a truth that’s sometimes overshadowed by hype over algorithms: a product that people love to use can beat a product that has a better engine under the hood but a poorer experience. Especially in AI, where many offerings appear magical at first, the long-term winners will be those that integrate into our lives or workflows most effectively. As product exec, James Colgan notes, “Every product works – the question is no longer can it be built, but how does it feel to use?”. If it feels great to use and keeps getting better tuned to me, I’m sticking with it – and that is a moat.

Conclusion: Merging Old and New Moats for AI Startups

The panel at StartX ultimately painted an encouraging picture for founders. Yes, the AI landscape is incredibly competitive and fast-moving – today’s novel feature could be table stakes tomorrow. But by combining classic defensibility with the system of engagement, startups can still build lasting companies. A savvy AI entrepreneur should still ask: Do we have proprietary data? Do we benefit from network effects? Can we leverage scale or partnerships or brand to our advantage? – all the traditional moat considerations. At the same time, they must also ask: Are we crafting a unique feedback loop and an experience that users can’t live without? How can our product get better with each user interaction, and how do we make users feel a personal loss if they stop using it? Or more importantly, if the user were to switch to a new AI, the user would feel like they have months of work in understanding a completely new system.

Answering those questions may involve questioning the nature of the product and the UX of the feedback loop, – another way of framing this, how is the startup having its product build a relationship, not just a building a product.

As the discussion highlighted, thinking in terms of engagement encourages founders to focus on delivering value in a deeply user-centric way. It shifts the mindset from “I built a powerful AI, job done” to “I’m continuously co-evolving with my users.” And ironically, that might be one of the most “old school” business truths of all – understanding your customer at a profound level is itself a competitive advantage. The twist in the AI era is that the products can now literally learn from and adapt to each customer – which the user is doing the same – at scale, creating a moat through intimate personalization and habit formation.

In the words of tech strategist Ben Thompson, "great companies often succeed by leveraging new technology while still obeying timeless business principles".

The emergent principle from this StartX panel is exactly that: durability comes from technology and the human touch. The winning AI startups will be those that marry cutting-edge models with engrossing user experiences. By doing so, they build moats that are not just about code or data – but about connection with their users. And that, ultimately, may prove the firmest foundation in the age of AI.

Sources:

1. Jerry Chen, “The New New Moats,” Greylock Partners – on how cloud and open-source are reshaping moats, and definitions of systems of record vs engagement
2. Sajal Sharma, “What’s the Moat? Product Defensibility for AI Applications” – discussing the challenges of defensibility in an era of foundation models and the importance of user experience and feedback loops
3. Anuj Shah, “Summary of 7 Powers (Hamilton Helmer)” – overview of classic business strategy moats like network effects, switching costs, scale economies, and brand
4. Crunchbase News, “AI Note-Taking Startup Abridge Raises $300M…” – example of an AI company integrating deeply into workflow (doctors) and its rapid adoption in healthcare
5. James Colgan, “Design Is the Ultimate Competitive Moat in B2B Software” – insights on how superior UX, onboarding, and emotional engagement drive product stickiness in the age of AI