The first version of Kognitos was hiding in five minutes of a child's handwriting. During the pandemic, Binny Gill's ten-year-old son tried to build tic-tac-toe in Python. He finished in a couple of days. Gill was proud, then unsettled. At roughly the same age, some thirty years earlier, he had made the same game in roughly the same time. Computers had become immensely faster. Programming had not become meaningfully easier.
Gill offered another exercise: write a program to tell whether a number is prime. The instruction to “think like a machine” went nowhere. So father and son switched to pseudocode - a plain-language account of what the program should do. That took minutes. Translating those notes into working Python took hours, until the younger Gill lost interest.
Then came the beginner's question that an expert might have been trained not to ask. Why couldn't the notes themselves work? Why couldn't programming be like talking to Alexa? Gill laughed, paused, and eventually told his son not to learn Python. The premise behind Kognitos had arrived: perhaps the burden of translation belonged to the computer.
“Why should humans need to think like machines when machines can now think like humans?”Binny Gill
A career spent removing machinery
That reaction did not come from a newcomer to software. Gill studied computer science and engineering at IIT Kanpur from 1995 to 1999, then completed a master's in computer science at the University of Illinois Urbana-Champaign. His 2001 thesis examined dynamic, policy-driven access control for computing environments. He went on to IBM Research, where storage systems became a long-running laboratory for the tradeoffs inside complex machines.
His name appears across work on caching, storage performance and reliability. By 2014, when he was writing about flash inside servers, he was already credited with more than 50 patents; recent biographies put his portfolio close to 100. The raw count is less revealing than the subject matter. Gill kept working on systems whose sophistication should disappear behind a simpler experience.
That instinct carried into an eight-year run at Nutanix. As chief architect, and later CTO for cloud services, he helped shape infrastructure meant to collapse separate storage, computing and virtualization chores into something easier to operate. At a 2017 data-center event, he compressed his taste into six words: “Complex is competent. Simple is genius.” It could serve as a caption for the rest of his career.
Gill left Nutanix in January 2021 to focus on the company he had founded the previous year. The move changed the surface area of the problem, from infrastructure used by specialists to everyday business processes managed by finance, operations and supply-chain teams. The design obsession remained: take a complicated technical layer and move it out of the user's way.
The translation tax
A business process usually has an unofficial owner. It may be an analyst who knows which purchase orders require a second look, a logistics coordinator who recognizes a missing field, or a finance lead who can explain why one invoice is normal and another is not. Traditional automation asks that expert to hand the knowledge to someone who can document it, model it, code it and maintain it.
Each handoff creates distance. The expert speaks in examples and exceptions. Software wants fields and rules. When the real world changes, the people closest to the work often cannot repair the automation themselves. Gill's label for the alternative is “English as code”: the process is written in readable language, executed by the system and refined through questions when an exception appears.
The old handoff, compressed
There is an important wrinkle. Natural language is expressive because it tolerates ambiguity; business software is dependable because it does not. Kognitos cannot simply put a chat box in front of a workflow and call the result programming. Gill's answer is a neurosymbolic architecture: neural models help interpret unstructured material and human intent, while a symbolic layer represents the explicit rules that execute.
Gill arrived at that architecture by revisiting an old lesson from computer-science theory. He had been taught that English is not a context-free grammar and therefore could not be parsed with the classic methods used for programming languages. While building Kognitos, he chose to encounter the limit himself. He began with traditional parser tools and found that local context could make portions of English behave deterministically enough for the job. The approach did not remove ambiguity. It made ambiguity visible, giving the interpreter a point at which to ask a person what a sentence meant.
That detail captures his first-principles temperament. Gill has said he needs to hit a theoretical wall himself before accepting that it is real. It also keeps the product's promise more grounded than “AI understands English.” The system does not need a complete philosophy of language. It needs to recognize the business concepts in front of it, preserve their relationships and decline to invent a rule when the words do not supply one.
The flexible side
Language models recognize patterns, interpret messy documents and help turn a person's explanation into structured intent.
The governed side
Symbolic logic follows the defined process, exposes the reasoning path and stops for guidance when instructions are incomplete.
The combination expresses Gill's broader skepticism about confident machines. A conventional program tends to crash when it reaches an unhandled state. A generative model can remain creative and persuasive on the error path. For an enterprise buyer, a successful demonstration shows that the model worked once. Gill argues that the more useful test is how it behaves when the obvious answer is unavailable.
That test changes what counts as a good interface. An error message written for a developer pushes the business expert out of the loop. A plain-language question can keep the person who owns the process involved. Once the expert resolves the exception, the approved guidance can be documented for the next occurrence. The automation gains capability without quietly rewriting its own instructions. Maintenance becomes a conversation with a record, rather than a ticket sent into a technical queue.
A machine that says “I need clarification” can be more valuable than one that always has an answer.
The steering wheel
Gill often reaches for an industrial-age metaphor. Machines became physically stronger than people, but operators retained a steering wheel. AI promises to relieve mental labor, yet its control surface is still being negotiated. In his view, ordinary language can become that steering wheel only when instructions remain readable, corrections require permission and the resulting process leaves an audit trail.
This is also why Kognitos talks about institutional memory. In exception-heavy operations, the real procedure rarely lives in a pristine manual. It lives in the accumulated judgment of experienced employees. When those people leave, the edge cases leave with them. A system that records a question, the human-approved response and the resulting process change can make that tacit knowledge durable.
The ambition is large but specific. Gill has said there may be about 30 million developers in the world, while a billion people can reason step by step about how to accomplish a task. He wants the second group to become builders without first joining the first. Kognitos is less interested in teaching everyone syntax than in promoting the explanation they already know how to give.
From a lesson to a platform
Capital has given the proposition room to grow. A $20 million Series A in 2023 brought Kognitos' reported total funding to $30 million at the time. In June 2025, the company announced a $25 million Series B led by Prosperity7 Ventures, alongside the launch of its neurosymbolic platform. The product was aimed at hundreds of back-office use cases across manufacturing, retail, logistics, finance and other operations where unstructured documents meet strict rules.
Partnerships with companies including Wipro have pushed the idea toward enterprise deployment. Kognitos describes a Process Refinement Engine that keeps workflows current as people resolve exceptions. Gill, meanwhile, has continued to sharpen the public argument. In 2026 he described the shift as a move beyond prompts and toward knowledge: not a clever instruction that disappears after an answer, but a governed process that remembers how a business decided to work.
The son's prime-number notes still make the cleanest model. They contained intent that a person could understand immediately. The difficult part was not the reasoning. It was the ritual of translation demanded by the computer. Gill's company is a wager that AI can finally absorb more of that ritual without hiding the rules from the humans responsible for the result.
There is a playful inversion in the outcome. The father began by asking a child to think like a machine. The child responded by asking why the machine could not meet him halfway. Gill took the question seriously enough to reorganize his career around it. In a field crowded with promises of autonomous intelligence, his most consequential product choice may be the humble follow-up question - the moment the machine admits that human language, like human work, needs a little more context.