Every time Anthropic drops a product update on X, a startup somewhere dies.
That’s the sentiment across the startup world right now. Claude can do new things each week that were once considered core products of companies, not just simple feature additions. Each week, founders watch their whole business become a footnote, and the post-mortems begin.
I recently spoke to Eyal Fisher, co-founder and CPO of Sweet Security, and former head of the cyber operations center in IDF Unit 8200, about this. Even though the concerns are real, the lesson is being misread.
When a major AI platform ships a capability that overlaps with an existing startup’s product, the instinct is to panic. Fisher pushes back on the reaction. “I think that in most cases, it’s a little bit too early,” he says. “If that capability will miss important things, really bad things can happen. So I’m not sure that everybody’s running to replace all the security tools with AI — yet this is a trend. This is where it’s going.”
After more than two decades in Israeli military cyber operations, he co-founded Sweet Security in 2022 alongside former IDF CISO Dror Kashti and Unit 81 veteran Orel Ben Ishay. The company has since raised $120 million, including a $75 million Series B led by Evolution Equity Partners, and grown its enterprise customer base tenfold. So when Fisher talks about what it takes to survive, he’s speaking from experience.
The Core Capability Trap
According to Fisher, founders who treat these announcements as an obituary are making a strategic error.
The deeper problem, he argues, is that founders are building companies around capabilities, rather than ecosystems. “Let’s say that you’re trying to invent some kind of system that can summarize calls,” he tells me. “That capability is a waste of time to develop today because AI is doing it like that.”
Companies that once developed transcription and call summarization as their core product a decade ago do face a reckoning: not because they ‘failed’, but because the ground shifted underneath them.
For example, I use a media platform to record my podcasts, and an instant transcription is available as an extra feature. I no longer need an entirely new service because it’s part of the suite I operate in. It’s great as a user, but Fisher said that founders need to consider this everywhere and build the moat around the capability, not inside it. “You need to make sure that it’s going to be easy to use, interact with everything else that you have in your company, have a full ecosystem.”
This pattern is already playing out across software: from media tools to sales platforms to developer products. In other words, the feature will be commoditized. But the platform and its integrations are harder to replicate.
What Every New Founder Should Know
Fisher is a founder with 25 years of experience behind him, as opposed to many founders who are only 25 years in age. His advice to young entrepreneurs starting today urges them to pursue things that AI cannot disrupt today.
“Go after things that cannot be disrupted today by AI,” he said. “You don’t want to build something that someone else can do exactly the same in like half a year… If you think that you can do something in half a year, someone else can do it in half a year as well.” But then came the truth underneath that advice: “At the end, AI will replace everything.”
For many users, including myself, features like transcription are now embedded into existing tools I use for video recording and editing - I’m not looking for a solo tool anymore. So, his advice is particularly relevant for founders thinking about what to build next.
If replacement is as inevitable as he says, then the question becomes how much time you have before it happens - and what you build around your core in the meantime. That way, you can avoid the dreaded update from a large AI giant that risks putting you out of business.
Sweet Security’s Own Answer
Fisher applies this logic to his own company. Sweet Security’s runtime sensor — the technical foundation of its cloud security platform — is written in Rust, a low-level programming language that makes it unusually difficult to replicate. “There is almost no other company out there that wrote such a sensor in that programming language,” Fisher says. “It’s very hard. It’s very complicated.”
But even he doesn’t treat that as a permanent shield. “Eventually, maybe it’s going to happen. Until that happens, what we are doing is building the ecosystem around it.”
The Bottom Line
For founders worried every time a major AI company posts on X, Fisher’s parting advice cuts through the noise: “Patience. One day you are here, one day you are here. You need patience and resilience. It’s a hard journey.”
The founders who survive the AI update cycle will be the ones who built the deepest and had the discipline to keep building when everyone else was busy panicking.
