6, July 2026
by: Grace McCarthy
Blog by Ramen Bhar, Principal Data Engineer, Arreoblue.
Part 1 went through all twenty-three things Databricks announced at Data + AI Summit 2026. This is the shorter, more opinionated version: the five I think are actually worth your attention, and why.
Status: private preview
Most customer data platforms share the same flaw: they copy your customer data out of your governed systems and into a separate marketing tool, where it goes stale and falls outside every rule you’d set up to protect it. CustomerLake’s pitch is that it doesn’t do that. Profile Agents build a single view of each customer directly inside the lakehouse; Campaign Agents run personalised campaigns from that same view, no export required.
It’s private preview, so early days, and it’s up against Salesforce, Adobe, and Segment, who’ve had this market largely to themselves. CustomerLake doesn’t have a long feature list to compete on yet. What it has is the architecture: no separate copy of customer data sitting outside your usual governance. That’s the appeal if you’re already committed to Databricks and don’t fancy exporting customer data into yet another system.
Status: released, generally available
There’s an abundance of tools that can answer a question these days. Genie One’s USP is that it doesn’t stop there. It uses connected tools on a schedule, across web, mobile, Slack, and Teams. In practice that means something that finishes a task and lets you know it’s done.
Two things to know before you get too enthusiastic. Genie is only as good as Genie Ontology, the context layer underneath it, so that’s the thing to get right first. And it’s pay-per-use, which is friendlier for rollout than a seat licence but means you’ll want Unity AI Gateway in place before usage gets away from you.
Status: LTAP coming soon; Lakehouse//RT in beta
I’m covering these two together because they solve the same problem from different ends. LTAP gets rid of the duplicate copy of data that apps and analytics have traditionally kept, each synced by a pipeline that tends to hold up right until it doesn’t. Lakehouse//RT, built on a new engine called Reyden, makes the remaining copy fast enough to serve in real time. Databricks says it’s up to 16x quicker than the dedicated real-time systems people usually bolt on alongside a lakehouse.
A dashboard that’s a few hours stale is a minor inconvenience. An agent acting on stale or duplicated data can make the wrong call entirely, and won’t necessarily know it has. This is the least proven architectural bet on the whole list, so allow twelve to eighteen months before relying on it, and test the speed claims against your own workloads.
Status: announced; Panther acquisition agreed
Security data is large, expensive to store, and almost always sits in its own separate system. Lakewatch brings it into the lakehouse instead, backed by Databricks’ agreed purchase of security company Panther, its third security acquisition, so this is clearly a sustained push, not a one-off. It also watches what your AI agents are doing, flagging things like personal data leaks.
It’s early, contingent on the Panther deal closing, and up against established security vendors. Most compelling if you’re centralised on Databricks and eyeing your security tooling bill with some resentment.
Status: open source and free; managed version in beta
Most companies experimenting with agents are already running several different frameworks at once with no single place to control any of it. Omnigent sits above all of them, lets you swap frameworks with a one-line change, and applies one set of spending limits across everything.
Matei Zaharia called it a “manager of managers.” It’s open source for the same reason Delta Lake, MLflow, and Unity Catalog were: give something away, let it become the standard, then sell the managed version to everyone who needs one. There was also a blunt warning at the summit that running agents “is going to get extremely expensive,” which is presumably why this exists.
Those are the five. Part 3 steps back from the individual announcements to look at the pattern across the last three summits: what Databricks is actually trying to become.
If you are interested in finding out more, please contact us.
Have a question or want to make an enquiry?
Contact us
