Ask a company how much revenue it made and you might expect a number. Ask which definition of revenue it used, which customer record it trusted and which exceptions it counted, and the conversation gets longer. For Assaf Henkin, that expanding conversation is where a business begins to become intelligible. It is also where an AI agent can get into trouble. A computer can retrieve a figure with great confidence while knowing remarkably little about what the figure means.
Henkin, the co-founder and CEO of Jedify, has spent years working on versions of this difficulty. At Kontera, the material was web content and the interests of the people consuming it. At Amobee, it was the relationship between those signals and a marketer’s decisions. Now it is the accumulated knowledge of an enterprise: definitions, documents, systems and the working rules that make them useful together.
The continuity is more interesting than a simple change of industry. His career keeps circling a question that sounds modest until someone tries to answer it in software: how do you turn information into an understanding of what is going on? The latest answer has attracted investment. The question has been following him for considerably longer.
Before the agents, there were the words
Henkin’s early work included serving as a technical lead on large e-commerce applications at iXL. His education combined a BS in Computer Information Systems from San Francisco State University, where he graduated magna cum laude, with the joint executive MBA program of Kellogg at Northwestern and Recanati at Tel Aviv University. Software and business were already sharing the timetable.
At Kontera, which he co-founded, he led product management and product operations. His own career biography described a network reaching more than 100 million unique visitors a month across over 15,000 publishers. Those are historical figures, belonging to that chapter of the business. They help explain the scale of the product problem: meaning had to survive contact with a great many pages and readers.
Kontera’s advertising platform linked words within web pages to external content. Its wider content-intelligence business examined what people consumed across the web, social platforms and mobile devices. The attraction for marketers was a better understanding of interests and attention. A word on a page was a clue; the useful work lay in understanding the circumstances around it.
In June 2014, Singtel’s Amobee acquired Kontera for $150 million. An acquisition gives a career a convenient punctuation mark. Henkin’s next work suggests a semicolon would be more appropriate. He continued inside the larger organization, applying content and audience intelligence to advertising decisions. The subject changed its letterhead before it changed its underlying concerns.
by Amobee
founded
Series A
A basketball game, seen from the ticket office
By 2017, Henkin was writing as Amobee’s chief insights officer. One of his articles began with the NBA playoffs and the abundance of three-point attempts. Basketball offered a familiar demonstration of analytics changing behavior. But his attention moved from what happened on the court to what happened around it: reaching fans, understanding audiences and selling seats.
The article argued for bringing different sets of audience information together. Sports franchises could learn from behavior across multiple points of contact and use that picture to make marketing more relevant. A fan was more than a single interaction. Seeing the connections mattered. The ticket office, it turned out, had its own version of the coach’s problem.
That piece supplies an unusually clear bridge to his later work. Years before Jedify’s context graphs, Henkin was discussing the value of unifying information so people could act on it. The comparison has limits: an advertising audience and an enterprise AI agent present different demands. Still, the recurring interest is visible. Collecting signals is the beginning of the job.
He subsequently co-founded Sproutt and served as its president and chief operating officer. That added another operating chapter between advertising technology and his current company. Across these roles, the practical setting kept changing. The recurring responsibility was to turn a technological proposition into something people could use.
Three founders, one familiar headache
Jedify was founded in 2023 by Henkin, Adi Elimelech and Erik Shani. The company places the launch of its Semantic Fusion technology in 2024. Those two dates describe different milestones: the formation of a business and the arrival of its platform. By then, its founders had already spent more than 15 years working together around data and intelligence.
In his June 2026 account of their reunion, Henkin described the recurring difficulty of extracting useful intelligence from fragmented information, both while growing a startup and after acquisition into a multinational organization. Language models gave the team a reason to revisit that frustration. They returned to a problem they knew through operating experience, with a new set of tools available.
The division of expertise is part of the story. Henkin’s writing about the company emphasizes Elimelech’s background in taxonomy and natural language processing. Taxonomy sounds like something kept in a museum drawer. In a business, the ability to classify things, distinguish meanings and connect related concepts can determine whether two teams are even discussing the same customer.
The founders’ stated ambition is to make AI applications understand the particular business using them. That emphasis on particularity matters. A general model may know a great deal about companies in the abstract. An individual company has its own habits of definition, exceptions and permissions. Henkin’s current work occupies the awkward distance between those two kinds of knowledge.
The business has its own dictionary
Jedify calls its underlying technology Semantic Fusion. The company describes a live context graph that connects structured information, such as warehouse tables and CRM records, with unstructured knowledge, such as documents and meeting records. Definitions, relationships and operational rules become part of the picture an AI application can consult.
A useful way to read that proposition is as an attempt to give software the background a colleague acquires over time. Knowing where a number lives is useful. Knowing which number the business means is useful in a different way. The distinction becomes consequential when software is expected to act rather than merely produce a plausible paragraph.
In September 2025, Henkin framed context as central to the effectiveness of AI agents. His argument put organizational knowledge at the center of their ability to reason and make decisions. It was an operating ambition as much as a technical one: systems would need to work with the meaning of the business, beyond the mechanics of retrieving its data.
An April 2025 collaboration with Firebolt showed another part of the approach. Jedify announced a private preview for connecting data and routing processing across cloud environments. The announcement described an e-commerce use case combining Salesforce, SAP and other information. Henkin also emphasized flexibility. A business that already has several systems is unlikely to greet another compulsory island with a marching band.
What the new money is meant to build
In June 2026, Jedify announced a $24 million Series A led by Norwest, with Snowflake Ventures, S Capital VC, Cerca Partners and Oceans Ventures participating. Norwest partner Assaf Harel joined the board. The financing followed an $8.5 million seed round in September 2023. Henkin’s announcement rounded the total raised to $33 million.
Those figures describe funding for Jedify. The editorial significance lies in what the backers believe needs building. Norwest’s investment argument emphasizes an independent, model-agnostic layer that can maintain business context. It points to Henkin’s Kontera experience and the team’s history with data pipelines as relevant preparation. There is an appealingly practical premise here: experience with the plumbing may matter when everyone wants a more intelligent tap.
By September 2026, Henkin was describing a further collaboration with Snowflake. Jedify’s context graph would help create and maintain governed Semantic Views for Cortex Analyst. The account addressed a problem any living business creates for a static model: rules and definitions change after the model has been shipped.
In that design, changes to business meaning can be tracked and carried into the relevant views, automatically or through approval. Verified questions and their SQL answers also become reusable knowledge. The detail is revealing. Henkin is discussing maintenance, revision and accumulated use. His ambition depends on a system continuing to understand the company on an ordinary Tuesday, after something has changed.
“Enterprise data is fragmented across systems, definitions, permissions, and workflows.”Assaf Henkin · June 2026
Patience, in a business that sells speed
Henkin’s advice to other founders has a quieter tempo than the market around enterprise AI. “Be patient,” he has said. He also recommends defining the problem, starting with a simple solution, building in stages and collaborating closely. These are instructions that leave room for learning. A startup cannot know all the useful answers before it starts.
There is a similar emphasis in how he describes the motivation for building: bringing capable, creative people together around consequential problems. His discussion of AI also includes privacy, accountability, human oversight and the continuing role of human ingenuity. The proposed autonomy comes with an acknowledgment that judgment remains part of the work.
His public writing offers a portrait of an operator interested in the distance between a promising technology and a dependable decision. The sports article, the content-intelligence business and the context graph belong to different moments. Read together, they suggest a durable curiosity about what information leaves out when it travels alone.
For Henkin, the next chapter rests on a fairly demanding promise: that software can learn enough of a company’s working meaning to become useful inside it. A business dictionary that keeps up with the business would be a worthwhile thing. Anyone who has sat through a meeting devoted to agreeing what a number means can appreciate the attraction.