Why AI feels generic: Replit CEO on slop, toys, and the missing ingredient of taste

There are a lot of ideas and experiments spreading in the world of AI right now. But as far as Replit CEO Amjad Massad is concerned, the results are unreliable, moderately effective, and mediocre.

“There’s a lot of similarity there,” Massad explains in a novel way. VB Beyond the Pilot Podcast“Everything looks the same, all the images, all the code, everything,”

it "slope," As it turns out, this is the result not only of lazy one-shot prompting, but of a lack of personal taste.

“The way to overcome the slump is for the platform to do more and for the platform’s developers to infuse the agent with flavor,” says Massad.

Listen and Subscribe beyond the pilot But spotify, Apple Or wherever you get your podcasts.

How Repliit Conquers Normal Toys

Replit tackles the problem of slope through a mix of special notation, classification features built into its design system, and proprietary RAG techniques. The team is also not hesitant in using more tokens; Masad notes that this results in higher-quality inputs.

Continuous testing is also important. After the first generation of an app, Massad’s team sends the results to a testing agent, who analyzes all of its features, then reports back to a coding agent about what worked (and what didn’t). “If you start testing in the loop, you can give feedback to the model and make the model reflect on its work,” says Massad.

Pitting models against each other is also an innovative strategy of Replit: testing agents can be built on one LLM, coding agents can be built on another LLM. This takes advantage of their different knowledge distributions. “This way the product you’re delivering to the customer is more labor-intensive and less messy,” Massad said. “You generate more diversity.”

Ultimately, he describes a “push and pull” between what the model can actually do and what teams need to build on top of it to add value. Plus, “if you want to move fast and you want to ship things, you have to throw in a lot of code,” he says.

Why is vibe coding the future?

There is still a lot of pessimism about AI, as Massad believes it does not live up to the hype. Chatbots are well established but they provide “minor improvements” in workflow.

Vibe coding is starting to take off in part because it’s the best way for companies to effectively adopt AI, he said. “It can make everyone in the enterprise a software engineer,” he says, allowing employees to solve problems and improve efficiency through automation, requiring less reliance on traditional SaaS tools.

“I would say that the population of professional developers who study computer science and train as developers will decline over time,” says Massad. On the other hand, the population of Vibe coders who can solve problems with software and agents will grow “tremendously” over time.

Ultimately, enterprises must fundamentally change the way they think about software; Massad argues that traditional roadmaps are no longer relevant. Because AI capabilities are evolving so dramatically, builders can only “roughly” guess what things might look like months or weeks into the future.

Reflecting this reality, the Replit team remains agile and does not hesitate to “drop everything” when a new model comes up to evaluate. “It will ebb and flow,” argues Massad. “You have to be very aware of it and not get cocky about it.”

Listen to the full podcast to hear about it:

• “Squishy” divisions in AI intelligence hinder specialization;

• The cathedral vs. markets debate in open source – and why a “cathedral made of markets” might be the best path to collective innovation;

• How Replit “forks” the development environment to create separate sandboxes for experimentation;

• Importance of reference compression;

• What really defines AI agents: They don’t just retrieve information; They operate autonomously, repeatedly, without human intervention.

Subscribe to Beyond the Pilot apple podcasts, spotify And youtube,

<a href