Technical moats for commodity software are gone. Marek Hazan’s Felt Sense deployed AI agents to autonomously rebuild startups from Y Combinator's Winter 2026 batch, finding 10-20% were trivial to replicate. On *This Week in Startups*, Jason Calacanis called it a necessary “bucket of cold water” for founders. If an AI can clone a product in an afternoon, defensibility must come from proprietary data, regulatory hurdles, or complex sales cycles - not code.
The replicability extends to entire creative and development workflows. On *The AI Daily Brief*, Nathaniel Whittemore used Google’s Notebook LM to generate “The Masked Medici,” a faceless YouTube channel with cinematic visuals, in hours. He then built a 1990s-style strategy game by exporting designs directly into Google AI Studio, where the AI acted as lead developer to fill functional gaps. The barrier to launching media and software products has effectively vanished.
For veteran builders, this isn't theoretical. On *No Solutions*, Bitcoin pioneer Martti Malmi stated he has stopped coding by hand, using agents like Claude Opus for a 10-100x productivity boost. He argues this finally lets small, decentralized teams compete with Big Tech on user experience. However, he also expressed deep concern about the economic value of human labor as agents take over creative and technical work.
The emerging model shifts founder advantage from fast execution to strategic direction and unique assets. Hazan's experiment shows the market splitting: you either control hard assets - data, hardware, regulated processes - or you are a clone-in-waiting. Meanwhile, agents like Bordy evolve from tools into principals with their own “taste,” managing social capital and even raising venture rounds autonomously.
This automation wave is cresting with both spectacular scale and imminent regulatory backlash. The telehealth provider Medvi, built by two people using over a dozen AI tools, is on track for $1.8 billion in revenue but faces an FDA probe for AI-generated fake doctors. As Calacanis noted, such cases define the peak of a bubble, where speed prioritizes substance until the system breaks.
The question is no longer if AI can build a startup, but what, if anything, remains uniquely human in the process.
Marek Hazan, This Week in Startups:
- We found that 10 to 20% of the batch was pretty highly replicable and was composed of basically the same sorts of components.
- Building agentic founders felt like something that people would not even be able to debate that AI can take your job.
Martti Malmi, No Solutions:
- How much do I still code by hand?
- Basically zero.


