Why your AI agent will fail (and how a ‘habitat’ can save it)

This article summarizes an episode of Y Combinator’s video series featuring Amjad Masad, CEO of Replit.

Amjad Masad, CEO of Replit / Photo credit: TED

A massive change is coming, says Replit CEO Amjad Masad. Soon, AI agents will allow anyone to create software just by asking for it. He predicts this will crash the value of today’s apps and spark a race for something new, as companies compete to build the online environment that these AIs need to function.

This shift will also flip the job market, rewarding people who can do many things over those who only do one thing well.

An AI’s “home” decides how useful it is

Masad believes building the infrastructure around an AI agent is more difficult than building the agent itself. He calls this environment the agent’s “habitat,” which must be secure and scalable to be useful.

Safety is a core requirement
Masad argues that the habitat must be “sandboxed,” or kept in a separate, protected area. He explains, “What you need is a virtual machine, ideally in the cloud, ideally not on your computer because [AI] agents can actually mess up your computer. They could do a lot of scary things.”

Scalability is non-negotiable
He also notes that the habitat “needs to be scalable.” This simply means that if a product becomes popular, its online home must be able to expand easily to support millions of people and work with any programming language or tool out there.

AIs need access, not just intelligence

The AI’s habitat has to be more than just a simple computer. It needs to give the AI agent the ability to automatically use the same tools a human developer would. To build and launch real software, AIs need access to things like website launching tools, information storage (databases), and systems for user logins and processing payments.

AIs need a rich toolset
Masad gives an example: AIs aren’t great at building login systems from scratch. It’s better to use a pre-built one. “With Replit,” he says, “with actually one line of code, we turn on authentication. It will basically just turn on a setting, and then you have user authentication.”

Agents need economic capabilities
He also argues that AIs will eventually need economic power, stating, “sometime in the future you would want your agent to have some kind of wallet to be able to go pay for services… A more radical idea is that your agent needs to be able to hire people. For example, if it hits a CAPCHA and it doesn’t know how to solve a CAPCHA, it should go and ask a human to go solve the CAPCHA for it.”

The value of software is disappearing

As AI makes creating software an instantaneous process, Masad predicts its economic value will collapse, rendering many current business models obsolete. He argues that in this new reality, simply having complex software will no longer be enough to protect a business.

Software will become dirt cheap
Masad predicts, “software will be dirt cheap, that no one will be making money on the traditional type of SaaS. I’m not saying this will happen tomorrow or even next year. I know it’s going to happen on the order of years.”

Custom-made software is the new normal

This drop in value is happening because even people with no technical skills can now create software whenever they need it. Masad shared a story about an HR employee at his company who built a custom organizational chart tool within few days. This shift could render many current SaaS models, where customers pay recurring fees for software access, obsolete within years.

Non-technical users can build powerful tools
Masad shares, “one of our colleagues at Replit, Kelsey, she works in HR. She’s never written a line of software in her life… [The existing options] were very expensive.”

AI is reversing an old trend

The future belongs to the independent creator