The founders are barely out of college. Both are still technically enrolled at IIT Delhi. Yet here they were in early February, posting LinkedIn promises that bordered on the reckless: 30% off for early customers, full refunds if we can't automate your workflow. Not exactly the cautious language you'd expect from a company going up against OpenAI, Anthropic, and Google.
RamAIn—yes, the capitalization is intentional—emerged from Y Combinator's most recent winter batch with a claim that, if true, would upend the nascent field of AI agents that control computers. Ten times faster than the incumbents, they say—though no independent verification has surfaced yet. Not 20% faster. Not marginally better. Ten times.
The assertion demands skepticism. It also demands attention.
At its core, the pitch addresses a problem that's more pervasive than most people outside enterprise IT realize. Vast swathes of business software never got the API treatment. ERP systems from the late '90s. Insurance carrier portals built in the Bush administration. Electronic health records that physicians loathe but hospitals can't afford to replace. These are the digital equivalents of geological strata—layer upon layer of legacy code that companies navigate through mouse clicks and keyboard shortcuts because there's simply no other way in.
"Most computer-use agents today are basically playing a really expensive game of Simon Says," explains Shourya Vir Jain, the CEO, describing the competition's approach. Take a screenshot. Feed it to a vision model. Decide what to click. Take another screenshot. Repeat. "It's slow, expensive, and brittle."
RamAIn's contrarian bet: pre-train agents on specific interfaces before deployment rather than figuring things out in real time. The system learns an application's structure and interaction patterns upfront—which buttons matter, where data entry fields live, how the navigation actually works. The company calls this "calibration," and they've built a demonstration environment dubbed Playgrounds to show it off.
Whether this approach actually delivers ten-fold speed improvements in production environments is, of course, the multi-million-dollar question. No independent benchmarks have surfaced yet comparing RamAIn head-to-head against, say, OpenAI's computer-use capabilities or Anthropic's recent Claude updates.
The Everything, Everywhere Problem
The technical scope is ambitious, perhaps more than the founders initially realized. Windows. macOS. Browsers. Desktop applications that predate the iPhone. Co-founder Vansh Ramani has been posting demos—Outlook navigation, WhatsApp Desktop automation—with visual overlays showing what the AI "sees" as it moves through interfaces.
Self-healing comes baked in, they claim. Pop-ups appear? Layout shifts? The agent adapts rather than breaks. Human oversight remains default, with real-time handoff capabilities when the AI encounters situations it can't handle. Full audit trails for compliance. And where modern APIs do exist, the agents will use them, blending traditional integration with UI automation.
The promised deployment timeline feels almost aggressive. "Just a few days" to train and deploy agents across a team's applications, according to launch materials on Y Combinator's platform. Challenging interfaces that need custom work? "MCPs and full agents" within a week, the company says.
That timeline, if accurate, would represent a dramatic improvement over traditional robotic process automation implementations, which can stretch for months.
Eight Verticals, Zero Named Customers
RamAIn has mapped out target markets with the precision of consultants—which makes sense, given Jain spent time at McKinsey before this venture. Procurement teams. Insurance brokers navigating carrier platforms. Healthcare providers drowning in EHR workflows. Revenue cycle management firms. Finance operations still running decades-old software they can't sunset. Retail operations. Pharmacy. Logistics.
The pattern across all eight: repetitive work trapped in systems that resist automation.
As of the most recent available information, though, no customers have been named publicly. The go-to-market strategy centers on design partnerships—early adopters who'll help refine the product in exchange for steep discounts and that headline-grabbing refund guarantee. No public pricing has been disclosed, which is typical for enterprise software at this stage but makes it difficult to assess where RamAIn might fit in corporate budgets.
The founders themselves present an interesting study in contrasts. Jain holds a FIDE chess rating of 2118—competitive but not quite grandmaster territory—and previously launched something called Genoshi. Ramani, the CTO, has the kind of academic resume that catches venture capitalists' attention: research at Carnegie Mellon's Machine Learning Department, work at the University of Copenhagen, publications in venues like ICLR 2025 and the American Chemical Society (2024). His contributions to nearest-neighbor search were actually merged into Meta's FAISS library, a detail that suggests genuine technical chops rather than just academic credentials.
Both are still pursuing degrees at IIT Delhi, that Indian engineering institution that's become something of a pipeline to Silicon Valley.
CB Insights lists a $500,000 convertible note with Y Combinator, which tracks with the accelerator's standard investment structure. The corporate entity—RAMAIN TECHNOLOGIES PRIVATE LIMITED—was incorporated in India in early February, with both founders listed as directors.
A Crowded Field, Getting More Crowded
Timing in tech is everything, and RamAIn's timing is... complicated.
OpenAI rolled out computer-use capabilities last year, touting benchmark performance on OSWorld and WebArena. Anthropic just announced major updates to Claude—including something called "Claude Cowork" and integrations spanning Gmail, Google Drive, Excel, PowerPoint. Google and DeepMind have been exploring agentic browsing through projects like Mariner.
The open-source world hasn't been sitting idle either. Skyvern, itself a Y Combinator company from an earlier batch, offers browser workflow automation using Playwright and vision models. Browser-Use, from the most recent winter batch alongside RamAIn, has racked up considerable GitHub activity. Projects like BrowserOS and HyperAgent provide frameworks for anyone wanting to build their own agentic automation.
Then there are the legacy RPA vendors—UiPath, Automation Anywhere—pivoting frantically toward what they're now calling "agentic automation." These companies have installed bases, sales teams, enterprise relationships. They're not going to cede this market without a fight.
What distinguishes RamAIn, according to Jain and Ramani, is the pre-training approach and the emphasis on desktop applications alongside browsers. Most competitors focus primarily on web automation. Whether that differentiation proves meaningful remains an open question.
The Proof Is in the Production
Industry commentary from February noted that only about 3.3% of enterprises currently use AI agents in any meaningful way, despite steadily improving benchmark scores. There's a chasm between demo and deployment, between controlled testing and chaotic production environments where every edge case you didn't anticipate will surface within the first week.
If RamAIn can actually deliver on its speed and reliability promises—and that's a substantial "if"—the addressable market is enormous. Every midsize company has workflows trapped in systems that don't talk to each other. Every back-office team has someone manually copying data between screens. The frustration is universal and expensive.
The refund guarantee certainly projects confidence. Maybe even overconfidence, though it's also smart marketing—removing risk for early adopters who might otherwise hesitate.
Now comes the unglamorous work: proving the technology holds up when the customer's ERP throws an unexpected error message at 2 AM. Demonstrating that pre-training actually scales across the wild variety of enterprise interfaces. Convincing IT departments to let AI agents loose in production systems where mistakes have consequences.
A two-person team from IIT Delhi versus OpenAI, Anthropic, and Google. The speed claims are bold. The timing is opportune.
Whether the execution matches the ambition? That story is still being written, one customer deployment at a time.
