Satya Nitta had a computer that could beat the champions of Jeopardy! The next assignment sounded almost reasonable: help it teach. At IBM Research, he spent years pursuing an artificial tutor. The awkward discovery was that knowing an answer and persuading a child to care about it were rather different accomplishments. A quiz show offers a scoreboard. A classroom offers thirty people with other things on their minds.
That disappointment became a productive turn in his career. Nitta went on to co-found Merlyn Mind, which put a voice assistant at the teacher’s disposal. More recently, at Emergence AI, he has pursued agents that carry out work across software systems. Across these ventures, a recurring question has become more exacting: what should a machine be trusted to do once it leaves the demonstration?
Before the classroom, there was empty space
Nitta’s route into artificial intelligence began inside the chip. He earned his doctorate in chemical engineering at Rensselaer Polytechnic Institute in 1998 and joined IBM in 1999. His early work concerned the connections between components in semiconductors, a place where the material between wires can matter almost as much as the wires themselves.
One answer involved air gaps. In the interconnect research he helped lead, carefully placed spaces reduced the unwanted electrical interaction between neighboring conductors. The apparent joke writes itself: a useful contribution to computing consisted of finding the right places to put nothing. The difficult part was manufacturing that nothing reliably.
The work earned recognition in 2008, including IEEE ACE’s Innovator of the Year award. It also required the less glamorous discipline of making a promising structure survive practical tests. A clever chip that cannot tolerate its surroundings remains an expensive conversation piece. That distinction between an elegant possibility and a dependable object would follow him into AI.
The tutor that changed the assignment
At IBM, Nitta moved from semiconductor technology toward cognitive computing and learning. By 2016, his work included the collaboration with Sesame Workshop. The possibilities went beyond a talking screen: his team was interested in physical blocks with digital signatures, tangible objects through which young children might interact with computing.
There was a serious ambition behind that range of experiments. A system that understood a learner’s language might adapt its responses to what the learner knew. Education offered an unusually demanding test of intelligence: understanding an answer would have to be followed by understanding another person.
The eventual reckoning was blunt. By 2017, the broader IBM tutoring effort had run out of road. Nitta later described a system that struggled to hold children’s interest and produced little discernible learning benefit. Being available for a conversation did not make the computer a compelling companion for one.
“At the heart of education, at the heart of any learning, is engagement.”
Satya Nitta
His objection became sharper as a new generation of chatbots revived the tutor dream. In January 2024, he publicly argued that generalized AI tutoring remained a distant prospect. He favored assistance for the people doing the teaching, with tools assigned specific jobs. This was a researcher narrowing his claim after a long encounter with the problem.
There is a difference between abandoning an experiment and allowing the experiment to alter the next one. Nitta’s subsequent companies make that distinction interesting. He kept the interest in language, computing and learning. The job description for the computer changed.
A teacher should be able to leave the laptop
Merlyn Mind, founded in 2018 with Ravi Kokku and Sharad Sundararajan, took an ordinary classroom inconvenience seriously. Teachers use digital materials while also trying to watch, question and respond to students. Operating the equipment can pull them back toward a computer just when they need to be somewhere else in the room.
The Merlyn assistant offered voice control over those digital tasks. Its early Symphony Classroom system combined an AI hub, microphones and a remote. The teacher could ask for material, navigate applications or control the screen while continuing to move around. The promise could be expressed without predicting the end of an occupation: let the teacher teach with fewer trips to the keyboard.
In a 2024 conversation with Michael Horn, Nitta described putting teachers first as a consequence of his IBM experience. The person who could engage a class was already present. Helping that person operate the technology offered a concrete place to begin. There is a pleasant lack of ceremony in that proposition. Sometimes progress means the adult can walk to the back row.
The company announced $29 million in funding when it emerged publicly in June 2021, led by Learn Capital. Its launch followed pilots in more than fifty classrooms across more than twenty schools. Those were early trials, rather than a measure of universal acceptance, but they put the product in the setting for which it was being built.
The room gets a vote
Voice computing has an inconvenient requirement: someone has to hear you. A classroom is an acoustic environment, a collection of devices and a moving conversation. In that setting, a polished response from an assistant matters only after the system has correctly understood the request.
Nitta had already made a related distinction in a 2020 discussion of speech technology. Consumer assistants were one application of the underlying technology. Learning could demand different uses and different development choices. Borrowing a familiar interface did not settle the question of what belonged behind it.
By 2023, he was discussing education-specific language models, with attention to privacy, efficiency and the information a system should use. His interest was in constraining a model’s work to a domain and a defined body of material. For a classroom tool, the freedom to produce an answer to almost anything creates its own engineering assignment.
These choices help explain the continuity between Nitta the chip researcher and Nitta the founder. Each environment imposes conditions that a demonstration can conceal. A circuit must survive manufacturing and use. A voice assistant must function amid actual classroom activity. A language model must give a response suitable for the task it has been assigned.
The signature on the sale
In August 2025, Nitta signed Merlyn Mind’s asset purchase agreement with Promethean as CEO. On September 9, Mynd.ai announced completion of the acquisition of software, patents, trademarks, know-how and other assets associated with Merlyn’s voice and remote-control technology. The transaction gave the classroom work a new corporate destination.
The buyer described plans to adapt and integrate that technology into Promethean ActivSuite and Explain Everything. It was an asset transaction with a defined technological scope. For Nitta’s story, the signature is also a useful date marker: he was still signing as Merlyn’s chief executive while his next venture had already become public.
Asset purchase agreement signed, then completion announced. Merlyn’s classroom control technology moves to Promethean.
From asking a question to getting something done
Nitta introduced Emergence in March 2024. His argument began with the improvement in computers’ understanding of language, then moved toward what remained difficult: planning, reasoning and acting. Better conversation opened a door. Getting useful work through that door required more machinery.
At AI Engineer World’s Fair that year, he presented the company’s agent approach. An orchestrator would coordinate other agents and models, while Agent-E worked through websites. The proposed workflow included planning, action and verification. A request could involve several steps and several systems, rather than a single generated answer.
Seen alongside Merlyn, the family resemblance is clear. A teacher asking for classroom material and an employee delegating a software task both want a computer to manage the intermediate operations. The user describes an intention. The engineering begins with everything that must happen afterward.
Verification gives that ambition a necessary complication. An assistant can report completion with excellent grammar while the underlying task remains unfinished. Once software is allowed to act, the quality of its explanation becomes only one part of judging it. The action itself needs inspection.
His interests also range beyond the next product. In a public biography, Nitta lists cosmology, astrobiology, nanoscale phenomena and the conservation of species and habitats. The spread is revealing without requiring a theory of his personality: he is interested in systems at very different scales, including some in which computing is hardly the main event. The biography also records his service on Rensselaer’s chemical and biological engineering advisory board, a continuing connection to the institution where he earned his doctorate.
What happens on the fifteenth day?
In June 2026, Nitta was among the authors of Emergence World, a research paper about evaluating agents over longer periods. Its platform placed populations of agents in shared simulated environments, with persistent memory, tools and mechanisms for collective decision-making. The study followed five worlds over fifteen days.
The paper reported sharply different outcomes under identical starting roles and conditions when different models powered the populations. This was a simulation study, with the scope and limits that description implies. Its useful question concerned duration: how does behavior change when an agent keeps operating and interacting, beyond a short, tidy assignment?
That question suits a career repeatedly interrupted by the demands of use. Nitta has spent time with systems whose promise is easy to admire in isolation. His more recent research asks what becomes visible when the observation lasts longer.
By October 2026, Emergence listed him as co-founder and executive chairman; he had also announced his role as chief scientist following Ian Eslick’s appointment as CEO. His stated focus included autonomous AI that could be verified, beginning with semiconductors. The subject brings him back toward the territory where his IBM career started.
Earlier in 2026, the company announced plans for a Bengaluru research lab, with an ambition to grow to roughly five hundred scientists and engineers over three to four years. That remains an expansion plan. It gives a sense of the scale of the research effort he wants to build, and of the distance between an intriguing agent and one that can be entrusted with consequential work.
The classroom lesson still travels well. A machine’s answer can be impressive, and the assignment can still be wrong. Nitta’s career offers the more useful possibility that a disappointing result can improve the next question. After all the work on making computers speak, he keeps returning to what deserves to happen when they do.