Databricks doubles down on agents
Databricks (businesswire.com) a partnership with Accenture to push enterprise agent apps, and also put a SOTA embedding model for agentic workflows into public preview on its platform (databricks.com). Those moves mean large embedding and inference workloads are moving to production — expect demand for high‑throughput GPU inference and optimized vector search stacks.
The Accenture Databricks Business Group will be supported) by more than 25,000 Databricks‑trained professionals, per the March 17, 2026 announcement. Clients named in the expansion—Albertsons, BASF and Kyowa Kirin International—are already working with the partners to build “agent‑ready” databases and AI applications on enterprise data, according to the press release. (businesswire.com) Databricks published Qwen3‑Embedding‑0.6B as a compact, 0.6‑billion‑parameter embedding model in public preview on Model Serving, explicitly billed as multilingual and optimized for retrieval. (databricks.com) Databricks says Qwen3‑Embedding‑0.6B is built on the Qwen3 foundation and comes from the same research team behind the GTE series, and it’s positioned for direct indexing with Databricks Vector Search. (databricks.com) Agent Bricks is described in Databricks product posts and docs as an automated system that "builds out the entire AI agent system," running additional methods and hyperparameter sweeps in the background to optimize models. (databricks.com) Databricks frames Model Serving as a secure, serverless deployment path and calls Qwen3‑Embedding‑0.6B “optimized for vector search and AI agent workloads,” which together point to productionized embedding and inference pipelines that will drive higher‑throughput inference and indexing needs. (databricks.com)
