Walking away from a reported $300,000 annual Meta job is the attention-grabbing part of Prakshal Jain’s startup story. The more consequential question is what he is building after leaving one of the world’s largest technology companies.
Jain and co-founder Alex Morris have launched Mitosis Labs, a startup betting that the next phase of artificial intelligence will be shaped not only by increasingly powerful models, but by the infrastructure that gives those models persistent memory, access to personal context and the ability to work across applications.
That is a crowded market, but the timing is significant. Global private AI investment reached $344.7 billion in 2025, more than doubling in a year.
The challenge for Mitosis is turning that enormous investor enthusiasm into a product people actually need and will pay for.
From Big Tech To AI
According to NDTV Profit, Jain spent nearly 3.5 years at Meta, working on frontend, web and extended reality experiences. His work included helping bring Facebook and Instagram to Meta Quest in 2024. The experience is relevant to Mitosis because the startup’s thesis is fundamentally about changing how people interact with technology.
Mitosis describes itself as a human-AI interface company. Its product stack includes Yappy, designed around conversational interaction, and Cortex, which the company positions as a private memory layer for AI.
The company’s website says Cortex can connect data from services such as Google Workspace, WhatsApp and GitHub, linking it into a knowledge graph that users and AI agents can query.
That points to a potentially important shift in the AI business. The race is no longer confined to who builds the strongest model. A growing layer of companies is competing to own the interface between users, their data and multiple AI systems.
What’s Mitosis About?
Mitosis’ central proposition is that users should not have to rebuild their digital context every time they change AI providers.
Its founders argue that personal memory should sit outside individual model platforms. In the company’s framing, Cortex is intended to function as a neutral layer that can work across models while allowing users to retain ownership of their accumulated context.
Mitosis says data is encrypted in transit and at rest, sources can be disconnected and purged, and ingested data is not used to train models. These are company-stated product and privacy claims, not independently audited findings.
The commercial model is already visible. Mitosis lists plans starting at $7.99 a month, suggesting it is testing whether consumers will pay directly for persistent AI memory and privacy rather than relying entirely on advertising or enterprise contracts.
Capital Is Getting Selective
The funding environment makes this a high-opportunity but high-pressure moment for an AI startup.
Stanford HAI’s 2026 AI Index found that US private AI investment reached $285.9 billion in 2025, compared with $12.4 billion in China. The US also recorded 1,953 newly funded AI companies. The scale of capital flowing into the sector creates opportunities for founders, but it also raises the competitive threshold.
India’s startup market presents a more nuanced picture. Funding fell 9% to $10.1 billion across 977 deals in FY26 from $11.3 billion across 1,020 deals in FY25. Yet technology startups raised $7.2 billion in the first half of 2026, even as the number of funding rounds fell 43% to 652. The pattern suggests capital is still available, but increasingly concentrated in companies investors believe can demonstrate scale or differentiation.
That matters for Mitosis. The company is entering a market where AI attracts substantial capital, but investors are not necessarily funding every AI idea equally.
Can AI Memory Scale?
The real test for Mitosis will be whether its technology becomes indispensable enough to survive the platform giants it is trying to sit between.
Google, Meta, OpenAI and other major AI companies already have access to increasingly broad user ecosystems. They can integrate email, documents, calendars and other services directly into their products. Mitosis’ counterargument is that users need a neutral memory layer that is portable across models and providers.
That is an ambitious proposition. The startup’s own research page currently reports a preliminary 91.7% end-to-end accuracy for Cortex on a sample of the LongMemEval benchmark, compared with figures it lists for competing systems. But Mitosis explicitly says the full 500-question evaluation is still underway, meaning the result should not yet be treated as definitive evidence of product superiority.
Jain’s departure from Meta, therefore, is less interesting as a story about quitting a lucrative job than as a bet on where value will accumulate next. If AI models become interchangeable commodities, the valuable layer could be the one that remembers the user, protects their context and moves with them.
Mitosis Labs is betting that this layer can become a business. The next few years will determine whether that is a new category or simply another feature that the technology giants eventually absorb.
Also Read: France Becomes the First EU Country to Ban Social Media Access for Children Under-15
