Building a Durable Business in the AI Era
Many young companies I met in the past year have adopted AI at some level, but many are also building for today’s market without accounting for what AI will make possible — or obsolete — in 12 months.
Building a durable business might have required thinking one step ahead, but thinking 5 steps ahead is now important. Today you can be a first-mover, and tomorrow you’ll find yourself competing with 10 other companies using AI to build, or being outright displaced by AI agents and LLM giants.
Straightforward optimizations (quicker, better, cheaper) are no longer sufficient. So what does a durable competitive advantage look like?
System vs tool
A tool does one thing and can be replaced. A system creates lock-in through interconnected workflows, data, and habits. The more embedded it is in users’ operations, the higher the threshold for change, regardless of what AI can do.
The test is switching cost. But rather than asking whether switching will be annoying, ask whether switching requires the user to rebuild something they’ve accumulated.
Take Runchise, which connects POS, inventory, and financial workflows into one system for Indonesian restaurant chains. A restaurant running on Runchise for two years has accumulated transaction history, inventory data, and staff workflows, all in one place. Switching isn’t simply annoying, but means rebuilding two years of operational memory from scratch. As AI improves, Runchise embeds it into the existing system, and the restaurant gets smarter without switching anything.
The feedback loop
“Data is key in the AI era” has become a default answer for founders asked about their differentiation. But the moat is in building a good feedback loop with the data.
The companies that ride on AI improvement are not just sitting on data, but building systems where every user interaction generates signal that makes the next interaction better, in ways specific to their context.
Rey, an Indonesian insurtech company, automates healthcare claims processing for insurers. Every claim it processes — treatment types, prescriptions, drug codes — gets contextualized against Indonesia’s specific healthcare system and provider network, and the more it processes, the better it gets at detecting fraud patterns and anomalies that only emerge at scale, in the Indonesian context. With hundreds of thousands of claims processed, Rey has accumulated edge cases, local coding inconsistencies, and insurer-specific rules that no general model will ever see. When a better base model ships, Rey fine-tunes on top of it with this loop data and immediately pulls ahead. The result: insurers achieve significantly lower claims ratios, which will continue to improve as AI gets better.
The moat is what the AI has learned from processing Indonesian healthcare claims at scale, not simply the data itself.
What AI can’t replicate
AI disrupts software capabilities faster than it disrupts trust. A founder who has built deep relationships with a specific user segment, channel, or geography has something that doesn’t transfer to a better model. For Indonesia, local context and trust are durable in a way that technical features are not.
Take Recompound, which manages equity portfolios for busy Indonesian professionals as their “Personal CIO”. The service is AI-augmented, but the durable part is the trust embedded in the client relationship. AI can replicate certain aspects of trust like reliability and transparency, but not accountability, aligned incentives, or the feeling of having someone genuinely responsible for the outcome. The service also removes the mental burden of active portfolio management for people who have neither the time nor the appetite for it. The durable advantage here is the relationship built over time that makes the advice trusted, not just the advice itself.
Not every business can ride on AI improvement. Some will just get disrupted. A simple diagnostic: “If the next frontier model ships tomorrow and it’s 10x better, does my business get stronger or weaker?”
If the answer is weaker — rethink.
. . .
Runchise, Rey, and Recompound are Genesia Ventures portfolio companies.