A General Catalyst company, run by a former Palantir SVP, that drops engineers and researchers inside enterprises and rebuilds the work from the studs. Anthropic and AWS are along for the ride.
Here is a thing that is true about enterprise AI, and that everyone in a large organization quietly knows: it is very easy to run a pilot and very hard to change how the company actually works. You can buy a model. You can put it in a slide. You can get a nice demo where the chatbot answers a question about your quarterly numbers. And then the pilot ends, the deck gets circulated, and the claims-processing department keeps doing exactly what it did in 2011. The demo was never the hard part. The hard part is everything after the demo.
Percepta is a company organized entirely around the hard part. It is a New York-based "Transformation Company" - a specific term of art at the venture firm General Catalyst, which formed Percepta, funds it, and owns it outright - and its model is refreshingly un-clever. Instead of selling you a point-solution product, or handing you a consulting deck and leaving, it deploys its own teams of researchers, engineers, and product managers directly inside your enterprise. They move in. They rewire the workflow. The frontier, in Percepta's telling, is not a model release; it is the moment a hospital billing process or a trading desk finally runs differently.
You will notice this is a slightly unusual thing to say about yourself. Most AI companies want to tell you the future is inevitable and they are it. Percepta's framing is more careful, and more honest: the prosperity is available but not guaranteed, and getting there is grinding, unglamorous work. That framing happens to be a very good sales pitch to a Fortune 500 executive who has already sat through eleven demos and has nothing to show a board for it.
Percepta's CEO and co-founder is Hirsh Jain, who spent seven years at Palantir Technologies, where he led the healthcare and civilian-government business and finished as a Senior Vice President. Before that: Google, Jane Street, the U.S. Department of Defense, and a machine-learning research paper or two, all after studying Computer Science and Mathematics at Harvard. This is a resume built specifically for the job of dragging a regulated institution into the AI era, which is presumably the point.
His co-founder is Hemant Taneja, the CEO of General Catalyst - which is how Percepta gets to be a wholly owned "Transformation Company" rather than a startup passing a hat. The rest of the founding bench is stacked in a deliberate way: CTO Thomas Mathew, Chief AI Officer Athul Paul Jacob, and Chief Scientist Costis Daskalakis - the last a celebrated MIT theoretical computer scientist - plus co-founders Radha Jain and Michael Rochlin. The team was pulled from Palantir, MIT, Meta FAIR, Google, Jane Street, Citadel, Scale AI, and Apple.
Enterprise AI is not one problem. It is a modeling problem, an engineering problem, and a product problem, all firing at once - and most teams show up with only one of the three.
Percepta's bet is that you need all three in the same room, at the same time, sitting inside the client. A research lab bolted to a field-engineering crew. That is the whole wager.
Percepta describes its work across three pillars - advancing the intelligence itself (foundation models, reinforcement learning, optimization), building the data infrastructure underneath, and getting the actual humans to adopt the thing. Here is what that looks like when it ships:
Cross-functional squads of researchers, engineers, and PMs deployed inside your enterprise to architect and ship change - not a product license, a resident team.
Led by Athul Paul Jacob and Costis Daskalakis, working reinforcement learning, foundation models, optimization, and operations research - then handing it to the field.
A client-intelligence platform built for Janus Henderson that helps distribution teams personalize outreach and read client needs across regions, running on Claude.
An AI-native research tool that synthesizes internal and external research so investment analysts surface market signal faster.
Clinical workflow automation, member engagement, revenue-cycle and utilization management, call-center transformation, and agentic operational command centers.
Modernized banking fraud detection, optimized manufacturing supply chains, and streamlined government licensing and permitting.
In June 2026, Janus Henderson - roughly $500 billion in assets, about 75 million clients - announced it had partnered with Percepta and Anthropic to build AI-native investment and client tools, and to roll out Claude Code and Cowork across the firm. The interesting part is not that a big asset manager is using AI; everyone is. The interesting part is the diagnosis underneath the deal: generic AI tools rarely align with an active manager's specialized workflows and proprietary data. Someone has to do the fitting. Percepta's argument is that it is the someone.
No seed-round theater. General Catalyst didn't write a check - it built the company and owns it. Revenue comes from long-term transformation partnerships, not per-seat SaaS.
The single question the whole thing rests on: can you rebuild an enterprise around intelligence before a competitor does?
Enterprise transformation is usually intense and, frankly, brutal - which is what makes Percepta's stated fifth value quietly radical. Culture here is not a perk; it is the thing that determines whether the engineers you embedded nine months ago are still standing. The full set:
The company runs on research-institute energy - offices in SoHo in New York and near Harvard Square in Boston, with London "coming soon." It is the kind of geography you choose when your hiring pool is equal parts frontier researchers and field engineers, and you want both to feel at home.
You generally do not file a lawsuit against a company you believe is irrelevant. In December 2025 - roughly two months after Percepta launched - Palantir sued former employees, alleging that Percepta had run an "aggressive campaign" to recruit its developers. Hirsh Jain had resigned from Palantir in August 2024. By January 2026, Percepta's executives were pushing back in court filings, characterizing the suit as a "scare tactic" meant to slow a young competitor down.
One does not need a view on the legal merits to notice the market signal. The AI-transformation space is now contested ground, and the thing being fought over is talent - the small population of people who can make a foundation model survive contact with a regulated enterprise. When the incumbent reaches for the courts against a two-month-old startup, the land grab has, at minimum, officially begun.
Interviews and product demos aren't hosted here - these links point to the primary sources where Percepta and its founders explain the model in their own words.