Why AI Seems Generic: Replit CEO Talks Dirt, Toys, and the Missing Ingredient of Taste

Currently, in the world of AI, there are many ideas and experiments circulating. But for Amjad Masad, CEO of Replit, the results are unreliable, ineffective and generic.

“There are a lot of similarities,” says Masad in a new VB Podcast Beyond the Pilot. “Everything looks the same, all the images, all the code, everything.”

This "slope," as it has come to be known, is not only the result of lazy one-shot instigation, but also a lack of individual flavor.

“The way to overcome the problem is for the platform to put in more effort and for the platform developers to imbue the flavor into the agent,” says Masad.

Listen and subscribe Beyond the pilot on Spotify, Apple or wherever you get your podcasts.

How Replit overcomes being generic

Replit tackles the slope problem with a mix of specialized prompts, classification features built into its design systems, and proprietary RAG techniques. The team is also not shy about using more tokens; this results in higher quality inputs, notes Masad.

Continuous testing is also essential. After the first generation of an application, Masad’s team passes the results to a testing agent, who analyzes all of its features, then reports back to a coding agent what worked (and what didn’t). “If you introduce testing into the loop, you can give feedback to the model and make it think about its work,” says Masad.

Pitting models against each other is also an innovative strategy at Replit: testing agents can be built on one LLM, coding agents on another. This capitalizes on their different distributions of knowledge. “That way, the product you deliver to the customer takes a lot of effort and is less sloppy,” Masad said. “You generate more variety.”

Ultimately, it 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 ship things, you have to throw away a lot of code,” he says.

Why vibe coding is the future

There is still a lot of frustration around AI because, Masad acknowledges, it doesn’t live up to the intense hype. Chatbots are well established but offer “marginal improvement” to workflows.

Vibe coding is starting to take off in part because it’s the best way for companies to adopt AI effectively, he notes. This can “make everyone in the company the 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 have studied computer science and trained to be developers will decrease over time,” says Masad. On the other hand, the population of mood coders capable of solving problems with software and agents will increase “dramatically” over time.

Ultimately, companies must fundamentally change the way they design software; traditional roadmaps are no longer relevant, says Masad. Because AI’s capabilities are evolving so dramatically, manufacturers can only “roughly” estimate what things might look like months or even weeks from now.

Reflecting this reality, the Replit team remains agile and does not hesitate to “drop everything” when a new model comes out to carry out evaluations. “There will be ebbs and flows,” says Masad. “You have to be very zen and not have any ego about it.”

Listen to the full podcast to learn more about:

• The “spongy” divide in AI intelligence that hinders specialization;

• The cathedral versus bazaar debate in open source – and why a “cathedral made of bazaars” may be the best path to collective innovation;

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

• The importance of context compression;

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

Subscribe to Beyond the Pilot at Apple Podcasts, Spotify And YouTube.

Source link