Hi all,

Today, we wanted the opportunity to share with our customers and the broader Data & AI community about our thesis on what we are building at Oxygen Intelligence, our core point of differentiation, and why we are building it now.

Why Now: Language Models as a disruptive force on enterprise data systems

For the last two decades, the phrase “Data is the new oil” has served as a common rallying cry for organizations to building robust internal data platforms to unlock business value from their data. We’ve come a long way since then. In particular, we’ve seen two substantial leaps in data:

1. The rise of elastic, cloud-native infrastructure that makes analytical computation efficient.
2. The development of frameworks that let teams build and productize data applications (dashboards, reports) and predictive models (classical ML algorithms).

SQL and Python became the lingua franca of this era. Still, in many ways, the innovation has been incremental.  And while the tools have improved, the main bottleneck remains the same: the technical know-how required to operate them—proficiency in programming, modeling, and managing complex data systems.

With the rise of Large Language Models in the 2020s, that technical bottleneck began to disappear. These models make it possible to build AI agents that can reliably translate between natural language L and programmatic output P with both high recall and high accuracy. In doing so, they bridge the long-standing gap between business intent and programmatic data queries. Perhaps even more profoundly, AI agents can now extend this translation beyond analytics: from product requirements written in natural language to the code that generates fully functional data products.

‍

‍

We now can live in a world where questions about enterprise data, and the productization of data applications, models and workflows can happen in the most ergonomic and democratic way possible: through natural language, in plain English. We can call this new paradigm Agentic Data Intelligence.

Data Agents and Workflows as an open, standalone category

The Modern Data Stack (“MDS”) has become the backbone of how organizations store, transform, and visualize data. In many ways, it feels like the natural home for data agents—after all, agents interact with the same components that power analytics today: the BI layer, the ELT layer, and the data warehouse.

However, the rise of data agents represents a more fundamental shift that necessitates a decoupling. While data agents can belong to an existing layer of the Modern Data Stack (”MDS”), e.g. as part of a BI tool, the ELT layer, or as part of the data warehouse, we believe that data agents, workflows, and the infrastructure powering them deserves to exist as a separate category. More strongly, not only as its own category, but on its on plane.

Firstly, it’s now well understood that separation of concerns across data systems enables independent scaling, more reliable feedback loops, and thus, better enterprise outcomes and lower Total Cost of Ownership (TCO). Examples of this shift include separation of storage and compute, the cleaving of the control plane from the data plane, and finally, the composable shape characteristic of the Modern Data Stack itself.

‍

‍

Secondly, making the system of record for agentic entities (such as agents and workflows) dependent on a specific vendor layer introduces risk through vendor lock-in. As the MDS continues to mature and consolidate, each layer is increasingly incentivized to become an “all-in-one” solution, replicating the closed-ecosystem dynamics that once characterized platforms like Qlik or Informatica. For enterprise buyers, this creates a brittle ecosystem — too many tools doing the same thing, none working well in concert.

The market instead needs an open, layer-agnostic, and extensible platform — one that serves as a neutral system of record for Data Agents and Workflows, integrating seamlessly across the entire stack rather than being captive to one part of it. This is what we aim to provide to the market at Oxygen Intelligence in our Oxy platform: an open-source Agentic Data Intelligence system that powers data agents, data workflows, and other data automations.

Determinism-first design

Unlike higher recall and lower precision environments that exist for agentic coding and agentic search, Agentic Data Intelligence operates in a lower recall and higher precision environment. This is one in which not answering a question is preferable to answering an analytics questions incorrectly. A number is simply wrong, not a tiny bit wrong, in stark contrast to a search result being a little off.

To that end, we are introducing the following concepts:

1/ Ontology infrastructure that provides both the semantic and operational definitions of how to acquire data and to operationalize workflows to get to and end work product

2/ Composable workflows that can be chained together in a reasoning chain by Data Agents to more deterministically get to most desired outputs.

‍

Accuracy-guaranteed Workflow Automation

Determinism-first design allows Oxygen Intelligence to work with our customers in a highly differentiated way. In our less than 1 year of existence, we have been working with high-growth startups as well as Fortune 1000 companies in helping them automate data analytics workflows with accuracy guarantees. This includes Q&A automation as an embedded analytics solution (embedded inside a SaaS product), internal sales & marketing Q&A deflection, as well as entire end-to-end document generation for executive reporting previously not possible through traditional BI and DWH systems.

We are discovering that in Oxy, we are building a truly platform product that can automate all types of data analytics workflows, including those that require multi-environment orchestration, i.e. one that requires chaining together different systems, including non-data systems.

Further, our code-native design allows a builder to maintain engineering best practices in maintaining the Ontology, Agents, and Workflows in a declarative fashion (in plain YAML), unlike in more brittle GUI-only systems. These two key features allow the Oxy platform to be both comprehensive and developer-focused.

A Call to Action

It is early days, but we believe that our determinism-first, open platform can become an important system of record for agents, workflows, and context needed to run both data-centric and generic workloads in an agentic fashion. We would love for you to join our journey in making this a reality.

Visit our website at: oxy.tech to book a demo or reach out to us at joseph@oxy.tech.

‍