What's still worth building?

How has AI changed the landscape?

AI has become increasingly more capable at writing code. Sure, it can still produce slop: software that works but isn’t secure, maintainable, or well optimized. And someone needs to notice and steer. But it is making producing and changing code cheaper.

This means implementation-based advantages are getting weaker. Things like:

• “My ideal customer is not tech-savvy, they won’t vibe code my product”: but a competitor could. And the entry barrier has lowered significantly, so that even less capable competitors can attempt building more complex software.
• “My code is so dope”: this advantage is getting smaller by the day. A huge feature list is easier to catch up to by a motivated competitor. A huge list of integrations, same thing. A very complex algorithm, that perhaps required years to achieve very high performance and reliability, AI can get close to in a lot less time. A hard to migrate from data model can be reverse engineered to be imported elsewhere much easier than before.
• “My UI is so dope”: a great UI still matters when customers work directly with your product. But if they increasingly delegate that work to agents using your API, your interface becomes a weaker reason to choose you over a competitor.

Meanwhile, you still need a customer who:

• Has a clear and expensive pain
• Can and is willing to spend money to fix it
• Can make the purchase decision, or has influence over who can

Customers also have another option: building the solution themselves. This was always an option to some, but AI makes that practical for more people. The term “vibe coded”, dreaded by so many product teams, is now a thing.

What hasn’t changed?

What is arguably the biggest advantage accessible to anyone building products: people buy from those they know, like, and trust.

In a world where AI can build software cheaper and faster than humans, we’re being hit by a wave of new software and pull requests by a bunch of unknown subjects. Who builds them? Does their software actually work? Will someone help me when it breaks? Will they still be around next year?

A competitor might reproduce the features that matter to your customers in a few weeks. But they still need to earn your customers’ trust that they’ll keep delivering, especially when things get tough. AI can help, but it can’t build a track record for you.

A strong brand helps people discover and remember you (the “know” and “like” part). Keeping commitments, communicating honestly, and showing up when things go wrong give them reasons to trust you. That takes consistency, and it’s especially difficult if you’re just starting out.

What other advantages remain?

There are a few other moats that are still somewhat safe:

• You hold the keys to your customers’ operations: you’re not just a tool they use, you are the authoritative resource for some of their key operations (e.g. who paid, who has access, what was agreed). AI may make moving the records easier, but replacing the processes and dependencies built around them is a different, much tougher problem.
• You can act on your customers’ behalf: you don’t just “know stuff”, you can “do stuff” with their authorization. Stuff like moving money, filing documents, deploying stuff. Having authorization alone isn’t much protection. Being trusted to do the job without someone checking every action is harder to replace.
• You’re approved by some governing entity: someone could vibe code what you do quickly, but you’re in a regulated environment where your credentials aren’t just an advantage, they are needed to operate. And those regulatory approvals can take a long time to get. That’s usually a big friction point that discourages competitors from getting into your field.
• Your customers get value from others being there: a competitor can recreate the software, but in order to reproduce the network effect, they need the people and organizations using it. AI doesn’t give them that.
• Your software gets better the more people use it: your software learns from usage patterns, and uses that learning to improve future results. AI can help a competitor copy the entire app, but still wouldn’t give them what it has learned.
• Running your software requires expensive infrastructure: AI can help you deal with the complexity of the infrastructure, but it can’t just print money you can use (wouldn’t it be nice, though?).

So what’s still worth building?

Products that solve problems people will pay to fix, built by people they trust or are willing to bet on.

AI has made it possible for more people to build software, to do so more quickly, and more affordably. It has made “this would be hard to build, not many will attempt it” a weaker reason to expect a lasting business.

AI hasn’t replaced the work of understanding your customers’ problems, showing up when they need you, and earning their trust.

Inspiration

These are some of the sources that inform my argument, and I recommend checking them out:

• Rob Walling, The Only SaaS Moats Left in 2026. Implementation-based advantages, customer trust, systems of record, authorization to act, regulatory barriers, network effects, and learning loops.
• Einar Vollset, SaaS Moats in the AI Era: Sell the Outcome, Not the Tool. Delivering outcomes and building learning loops that competitors cannot reproduce simply by copying the app.
• Anthony Eden, AI Generated Apps vs SaaS: Which Moats Still Work in 2026. Collaboration, regulated environments, operational excellence, and capital-intensive infrastructure.
• Rob Walling and Einar Vollset, Startups for the Rest of Us, episode 855: The Eight SaaS Moats that Still Work in the Age of AI. Customer trust and the eight-moat framework, including the distinction between moving data and replacing operational dependencies.