Databricks Just Shipped General AI Agents for Businesses
Databricks just launched general-purpose AI agents for businesses at its Data + AI Summit. The headline product is Genie One — described by CEO Ali Ghodsi as an agent that “computes whereas other agents recite.”
That sounds like marketing copy. But the architecture behind it is actually interesting, and it tells you where the agent wars are really being fought right now.
It’s not the models. It’s the context.
Genie One: an agent that works inside your data
Genie One is Databricks’ answer to the enterprise agent question: how do you build an AI agent that actually knows your business data without a three-month RAG implementation?
The agent is grounded in the Lakehouse — Databricks’ unified data and analytics platform — and uses what they call Genie Ontology to understand the relationships between your data assets.
Genie Ontology is a live context layer that continuously learns from internal and external data sources. It uses a Google PageRank-like algorithm to surface relevance, meaning the agent doesn’t just retrieve data — it understands which data matters for a given question.
The distinction between “recites” and “computes” is worth thinking about. A reciting agent paraphrases what it finds. A computing agent performs analysis, joins datasets, runs aggregations, and produces answers that require reasoning over structured data. Genie One is positioned as the latter.
The architecture: Lake TAP and Reyden
Behind Genie One sits two new infrastructure components that are worth understanding separately.
Lake Transactional/Analytical Processing (TAP) is a new architecture that lets AI agents access both operational and analytics workloads on a primary data lake copy in open format. This means agents can observe production databases, reason across datasets, and take autonomous action — all without the data movement overhead that typically bottlenecks agent deployments.
Reyden Compute Engine (named after co-founder Reynold Xin) powers the real-time Lakehouse with millisecond query latency for tens of thousands of concurrent users and agents. This is the infrastructure problem that most agent platforms hit: the model is fast, but the data retrieval isn’t. Reyden is Databricks’ answer to that latency gap.
The governance layer: Unity AI Gateway and Lakewatch
Databricks didn’t just ship an agent. They shipped the infrastructure around it.
Unity AI Gateway provides security controls, cost management, and agent monitoring for enterprise AI deployments. This is the governance layer that enterprise procurement teams actually care about.
Lakewatch is an agentic SIEM solution for deploying defensive security agents and automating threat detection. Databricks acquired Panther Labs earlier in 2026 for this exact purpose — an AI security operations center platform.
The combination is telling. Databricks is positioning itself not just as the platform where agents live, but as the platform where agent security is managed. This is a direct response to the BadHost vulnerabilities and vm2 sandbox escapes that have been surfacing all year. The security problem isn’t theoretical anymore.
The numbers
Databricks’ annualized revenue hit $6.9 billion at the summit, representing over 80% year-over-year growth. The company is in a rare position: growing fast while the enterprise AI market is still figuring out what it actually needs.
The top enterprise concerns, according to Databricks, are cost (AI token usage has skyrocketed), security (rapid AI expansion demands automated threat detection), and context (enterprises struggle to feed accurate, real-time data into agents).
Genie One is the answer to the third problem. Unity AI Gateway and Lakewatch address the first two.
What this means
The AI agent industry is at an inflection point. The first wave was about building agents that could answer questions. The second wave — which Databricks is declaring with this launch — is about building agents that can act inside your production data infrastructure.
The tension here is real. Ali Ghodsi said at the summit: “We believe that AGI is already here. AI does not have an intelligence problem right now. It’s plenty smart. The problem is that AGI is not completely permeating our organizations.”
The question isn’t whether agents are smart enough. It’s whether your data infrastructure is ready for them.
Lake TAP, Reyden, and Genie Ontology are Databricks’ answer to that readiness gap. They’re betting that the bottleneck isn’t model capability — it’s context delivery.
That’s a defensible bet. The companies that figure out how to reliably feed real-time, accurate business context into agents will win. The ones that don’t will have the smartest models in the world with nothing useful to do.
The infrastructure layer is where the real work is happening now. Databricks just made it more visible.