AI’s “second wave” redefines startups with new products

For the past three years, AI has primarily been used as a cost-cutting tool. More founders and investors are looking to move beyond that era.

They argue that the next chapter of AI will be defined by new types of products, including apps, games, companions, and services that could not exist before large-scale language models. They call it the “second wave” of AI.

“The first wave of AI made things cheaper, more automated, more efficient,” said Kailan Gibbs, a former Google DeepMind product manager who runs AI startup Inworld. “The next wave will create things that didn’t exist before: new products, new experiences, new revenue. That’s the difference between optimizing spend and creating it.”

For Gibbs, the difference is existential. If AI simply reduces costs, it will also reshape value within existing businesses. It expands the economic pie when it enables entirely new consumer products, products that people can pay to buy.

“AI reaches its true economic potential when it creates value that consumers want to pay for, not just value that businesses want to save,” he wrote on LinkedIn. That next step, he says, will require a new “consumer-scale AI stack” with real-time responses of less than 300 milliseconds, support for millions of users simultaneously, and highly personalized experiences tailored to individual preferences.

In January, Gibbs launched a Silicon Valley accelerator to support up to 30 “second wave” AI startups, companies that build new consumer experiences rather than bolt chatbots into old workflows. Venture capital firms like Khosla Ventures and Lightspeed Venture Partners are participating, along with leaders from OpenAI, Google, and Stripe. Demo Day will be held in San Francisco in early March.

This philosophy echoes a recent post from Y Combinator CEO Garry Tan. “Instead of worrying about doing the same thing for less, why not focus on something you never dreamed of doing?”

A handful of startups are already embodying that spirit.

Particle brings news for the age of AI

Sara Beykpour, CEO and co-founder of Particle, says the technology industry is at a breaking point.

“We are in a transition period between the first and second waves,” she said.

In the first wave, productivity increased significantly. Particle, an AI-native news platform, lets you build, test, and deploy tasks in hours that once took months.

“We actually call each other out in meetings when someone falls into an old way of thinking,” Beikpour said. “We jokingly call it ‘boomer thinking,’ even though we are all millennials.”

This mindset shift gives startups more time to focus on unlocking new AI-powered formats. Particle recently launched Podcast Clips, a feature that embeds the most relevant snippets of long-form podcasts directly into news articles. Rather than searching for three-hour episodes, users can watch curated clips related to specific topics.

“It changes the information hierarchy,” Beykpour said. “Rather than searching for the podcast you want to listen to, we deliver podcasts based on the most relevant parts.”

Under the hood, the system uses AI embeddings to map relationships between transcripts and stories. Talk show clips and President Donald Trump’s comments about Greenland and Davos can be automatically linked to related reports. Generative AI then overlays a summary and context on top.

These AI embeddings are “significantly improved in important ways,” Beikpour told Business Insider in a recent interview.

AI could become a “supermotivator” in fitness

If Particle reimagines news, Luvu uses generative AI to reinvent personal training.

Launched in August 2025 by CEO Alexis Sursock and CTO Creston Brooks, the AI-powered fitness app has already attracted approximately 250,000 users. The app is powered by an AI “Marshmallow” that acts like a personal trainer, sending highly personalized notifications and real-time feedback.

“The key is personalization, which is powered by AI models and was not possible until the advent of this technology in recent years,” Brooks said.

Instead of a generic reminder, “It’s time to train,” Luvu customizes the message. If a user records that they took a test yesterday, the app might follow up with, “Test finished, time for training!”

The results were amazing. Luvu’s notification click rate is 4x higher than a typical non-personalized prompt. Brooks claims that Luvu’s retention rate is two to three times higher in an industry where only 2% to 3% of users are active after 30 days.

The app offers three motivational styles: supportive, neutral, or “mean marshmallow.” Behind the scenes, Luvu also uses AI for fine-grained 1:1 messaging created by LLM.

The company is also experimenting with reinforcement learning using validated rewards, a relatively new technique for training and improving AI models.

Users can record their exercise by leaning their phone against a surface. The app uses computer vision models to verify whether squats and other movements are performed correctly and provides real-time corrections, such as “straighten your knees.” These validated signals are fed back into the system and help train what Brooks envisions as future “super motivator” models.

This is more than just a chatbot layered on top of your fitness tracker. This is a feedback loop between human behavior and AI that simply could not exist before modern models.

Status supports your dream life online

For Fai Nur, CEO and co-founder of Status, an AI-powered social simulation game, the second wave is about imagination.

“Before the LLM, this status could not have existed,” she said.

With over 3 million downloads, the app allows users to role-play in an AI-generated social media world. Think of The Sims as a living social feed.

Users can cast themselves as Hogwarts students, soccer stars, Stranger Things characters, or anything else they can imagine. Post an update and an AI-generated character will respond instantly. Events unfold dynamically. Missing a penalty kick or facing repercussions. The AI ​​system evaluates answers by assigning an “aura score” to level players up or down.

The non-deterministic nature of LLM output is problematic in many corporate environments. Generative AI models may respond to the same prompt in different ways, which is not suitable for applications that require exact accuracy.

In games, this can be an asset, as each new response the AI ​​generates is new, creating a richer and more diverse experience.

“We’ve never been able to role-play like this before,” Noor said.

Before LLM, creating an immersive fandom world required convincing other humans to join. Now, the entire social universe spins instantly.

For Gibbs and other second-wave AI proponents, that’s the point. The future of this technology will not be defined by incremental cost reductions, but by products that feel native to AI: experiences that surprise, motivate, inform, and entertain at the consumer scale.

If the first wave made businesses leaner, the second may make everyday life stranger, richer, and more interactive. And importantly, people will pay for it.

Sign up for BI’s Tech Memo newsletter here. Please contact us by email. abarr@businessinsider.com. Note: Axel Springer, Business Insider’s parent company, is an investor in Particle.