"Microsoft has announced the acquisition of Osmos, a Seattle-based startup specializing in data ingestion via AI agents. The deal is intended to strengthen Microsoft Fabric with autonomous agents that automatically prepare data for use in a wide range of applications. Osmos focuses on automating ETL (Extract, Transform, Load) processes that normally require a lot of manual work. Where conventional tools require rigid schedules, Osmos uses AI models to dynamically learn and transform data structures."
"At its core, Osmos revolves around two AI agents. The AI Data Wrangler automatically normalizes unstructured data (also referred to as "weird" or "messy" data by Osmos), whether it's nested JSON, TXT files, irregular CSVs, or PDFs. The agent derives relationships between source and target schemas without explicit rule coding. It relies on machine learning that is already much older and more familiar than the generative and agentic AI of 2026."
"In addition, the AI Data Engineer generates production-ready PySpark code for building pipelines. This agent handles complex logic such as multi-file joins or ERP migrations. The output is natively compatible with data lake environments. According to Microsoft, Osmos specifically solves the problem of reading external data from customers, partners, or suppliers where file formats are inconsistent and error-prone."
Microsoft acquired Osmos to integrate autonomous AI agents into Microsoft Fabric, improving automated ingestion and preparation of external data. Osmos automates ETL work that often requires manual effort by dynamically learning and transforming diverse, irregular formats such as nested JSON, text, CSVs, and PDFs. Two core agents power Osmos: an AI Data Wrangler that normalizes messy unstructured inputs and infers schema relationships without hand-coded rules, and an AI Data Engineer that generates production-ready PySpark pipelines compatible with data lake environments. Osmos targets inconsistent external data from customers, partners, and suppliers, addressing error-prone file formats. The acquisition details were not disclosed.
Read at Techzine Global
Unable to calculate read time
Collection
[
|
...
]