A corporate AI strategy can look wonderfully coherent in a slide deck. Then somebody opens a browser. One employee tries ChatGPT. Another buys Cursor. A team adds Copilot. A department quietly experiments with Gemini. The official list says ten tools; the working day may contain four times as many. Naman Ambavi has decided to build his company in that discrepancy.
Oximy, the San Francisco startup he founded in 2025, describes itself as a control tower for AI at work. The metaphor is revealing. A control tower does not invent the aircraft or tell passengers where to travel. It makes existing traffic legible enough to coordinate. Oximy is meant to show companies which AI tools are being used, what they cost, where information is moving and how work can be routed to the right model.
Ambavi is 23, an age that makes enterprise governance sound almost comically premature. He has addressed the obvious tension himself: “I am 23. I am building an enterprise security company. I know how that sounds. But the problem does not care how old you are. It just needs to be solved.” The line works because it acknowledges the eyebrow before returning to the work.
The recurring gap Ambavi described on the START podcast: what an enterprise believes it uses versus what a closer count can reveal.
Chapter one
The respectable value of an embarrassing first draft
Ambavi's path did not begin with a governance thesis. It began with customized phone cases. In 2018, he built PrintsMan, a no-inventory ecommerce shop in the classic dropshipping mold. The store failed. He keeps it in his public work history anyway, smiley face included, because he managed to sell the website to someone in Kolkata for $500. He calls it his first internet buck.
The sum was modest; the evidence mattered. A stranger would pay for something he had assembled online. At 16, web development felt to him like “the coolest thing ever discovered,” so he turned that fascination into UNDEV. Without the money or bandwidth for a conventional creative agency, he coordinated developers, designers, search specialists and writers around the world. Clients met one brand; a distributed network did the work behind it. Ambavi reports that the model handled more than $75,000 in portfolio value on Upwork and Fiverr.
This early chapter established a pattern: find people who can do the work, make the arrangement comprehensible to a customer, and learn what breaks once the polished front page meets delivery. It was improvised, slightly audacious and operationally educational. Plenty of founders erase the awkward prototype from the legend. Ambavi treats his as useful provenance.
Chapter two
Software meets the untidy world
In 2019, Ambavi co-founded Easocare. The company first built a personal records app, then changed direction and became a software provider for pharmacies and distributors in India. It eventually had an office in Mumbai, a team of 13 and two co-founders alongside Ambavi, Yaaresh Ferzandi and Aditi Ravani.
The numbers in Ambavi's account suggest how quickly a neat product idea became an operating system with consequences: an encrypted mobile app passed 10,000 downloads; enterprise deployments supported more than $400,000 in gross merchandise value; a logistics platform processed an average of over 50,000 orders each month. The company also managed a live public information tracker that drew more than 10 million users in four months.
“Easocare was a learning. A learning to scale things that don't scale typically.”Naman Ambavi, reflecting on the company
His retrospective is strikingly free of victory-lap polish. Easocare taught him partnerships, hiring, customer conversations and the constant negotiation between development and sales. It also taught the lesson that recurs throughout his career: the system drawn on the whiteboard is only an invitation for reality to object.
That same appetite for the operational layer appears in his volunteer work with Shrimad Rajchandra Mission Dharampur, which began in 2017. Ambavi says he has worked across technology, management, AI and events. He led interface work spanning more than 400 screens, built WhatsApp bots used by millions and helped with programs that brought 10,000 to 15,000 people together at once. More recently, the group's technical work has included real-time voice translation from Gujarati into six languages.
This is not the resume of a person who regards software as a clean object. It is software tangled with queues, volunteers, languages, logistics and hurried users. The tangles are the point.
Chapter three
A coffee, a flight and the hidden work of AI
Ambavi had known Aryan Sharma and Ayush Pathak through Twitter for more than four years before they worked together. As he wound down Easocare in 2023, a casual conversation with Sharma became a coffee, then a flight to Bangalore. Ambavi joined Induced as the first team member after the two founders.
Induced was building AI systems that could act on unstructured web data. Ambavi's remit spread across enterprise deployments, product work and human-in-the-loop operations. He describes a period of rapid experiments with large language models, generative interfaces, retrieval systems and open-source model fine-tuning. The company grew from a house in Bangalore to a presence in the United States. Ambavi secured an O-1 visa and took a one-way ticket to San Francisco.
The important discovery arrived between the model and the enterprise. Large companies wanted AI, but their approved-tool inventories were often theater by arithmetic. Teams were already using dozens of products across browser and desktop applications. Costs fragmented across departments. Policies existed without a complete view of what they governed. Well-intentioned people chose convenient tools to finish a task, and the behavior moved faster than the committee.
Discovery
Which AI products are actually in use, including the ones absent from the approved list?
Spend
Where do licenses overlap, sit idle or scatter across teams and cost centers?
Control
How can a company set policy and route work without pretending adoption can be reversed?
Oximy came from that observation. Its approach sits at the network layer, where it can identify AI activity without waiting for a separate integration with every new vendor. The company organizes the view around discovery, spend intelligence and control. When another AI tool appears next month, the product thesis says its traffic should already be visible.
“You can't govern what you can't see.”Naman Ambavi, Oximy's launch note
There is restraint hidden inside the slogan. The first enterprise instinct is often to ban whatever cannot be controlled. Ambavi argues that blanket restrictions invite workarounds because employees are usually chasing speed, not sabotage. The more useful question is how to permit AI without surrendering visibility. Oximy wants to be the instrument panel for that compromise.
The current bet
The traffic is already in the air
Oximy joined Y Combinator's Winter 2026 batch. Ambavi has described the three months as intense and clarifying, a compressed course in thinking, building and selling faster. The company emerged from the batch with the vocabulary of a founder who has learned to make an enterprise product understandable in one breath: see every AI tool and dollar, run work in an app the company owns, and route each request to the right model.
In August, he brought the control-tower argument to an Ai4 stage in Las Vegas. His recent writing has pushed on another weak point in corporate AI: businesses can usually describe their spending, yet struggle to measure what the spending changed. After conversations with more than 75 companies, he wrote that AI often has budget, momentum and executive attention but lacks an owner, a baseline and a way to prove the outcome.
That line widens Oximy's ambition. Counting tools is the entry point. The more consequential question is whether a company can connect AI activity to useful work while retaining control. Visibility becomes a prerequisite for measurement, and measurement becomes the defense against both waste and panic.
Ambavi's personal rules fit the company he has chosen to build. He values rapid execution, high enthusiasm and learning from direct experience. He quotes Reid Hoffman's line about a network being one's net worth, then translates it into a preference for people who are smarter, challenging and willing to help him grow. He has traveled to more than 23 countries and credits those encounters with shaping his thinking.
The connection can be drawn too neatly, so it is worth leaving some rough edges. Oximy is young. Enterprise infrastructure takes time, trust and patient selling. A compelling metaphor does not guarantee a durable system. Yet Ambavi's prior work has trained him for precisely the portion of company-building that refuses to fit into a pitch: deployments, coordination, iteration and the daily correction of an elegant assumption.
His stated plan is pleasingly plain: “stay close to customers and let the work speak for itself.” It is less theatrical than promising to transform the enterprise. It is also harder to fake. AI adoption has already escaped the official spreadsheet. If Oximy succeeds, it will be because Ambavi and his team made that unruly reality visible, useful and governable, one previously uncounted tool at a time.