Coming back from being at Data & AI Summit, our team is blown away by the amount of excitement being created by Databricks and the opportunity in front of us and our clients to create organizational efficiencies and customer experiences that would have seemed impossible even just a few years ago. It’s been said by many before, but at this time in the digital transformation wave that generative AI is driving, it’s not the models your organization is using or your ability to simply create agents that differentiates you from the pack. Access to frontier models, tooling and skills to create agents – those capabilities are simply table stakes in 2026. What differentiates organizations’ ability to move AI from simply providing incremental improvements in personal productivity to actually being a transformative technology that changes how and by whom works gets done and experiences are delivered to clients is their ability to provide agents with trusted and timely context and tools in a governed manner that enables agents to safely work with knowledge of your business, your customers, your products, etc.
If you were at Data & AI Summit, it’s clear Databricks is raising the bar for the rest of the industry - and doing it on two fronts. First, they’re building a cohesive platform for your organization to develop a rich, governed ecosystem of data & AI assets that give agents the context they need to do truly useful work. Second, they’re collapsing the historically siloed application and analytical layers of an organization - something they started last year with Lakebase and Apps - to bring operations and analytical intelligence closer together, so operational data feeds seamlessly into your analytics and vice versa. What that unlocks is smarter operations: people and agents making real-time, data-driven decisions in ways that, only a few years ago, felt reserved for the world’s largest, most capable technology companies like Uber, Netflix, and Google. Let’s break down some of the week’s biggest announcements:
Omnigent
Omnigent is an open-source meta harness for your agents - a single, governed entry point that lets you direct Claude Code, Codex, Cursor, and the rest in concert rather than one at a time. Each coding agent brings its own specialty, and Omnigent coordinates the work between them so the whole becomes greater than the sum of its parts. Run it on a server and the coordination comes under policy: you can maintain a central repository of skills, prompts, and instructions that apply to every agent across your organization, so your guardrails and institutional know-how live in one place instead of scattered across individual setups. It also handles the practical realities of running agents at scale, from cost management (when one provider’s session usage runs out, a policy can switch seamlessly to another) to collaboration (multiple users can work in the same coding session when a decision needs more than one set of eyes before continuing).
And because the details that delight still matter: it has a lovely little starfish logo.
More Genie with More Capability
Genie is Databricks’ natural-language interface to your data, and at DAIS it granted us two big wishes. First, the whole Genie family now runs on Genie Ontology - the piece we’re most excited to dig into at Origin. Anyone who’s watched an agent answer confidently and wrongly knows the problem is rarely the model; it’s that the model doesn’t know what your data means in the context of your business. Ontology reads across your tables, queries, dashboards, pipelines, and the 50-plus apps your teams already use (Slack, Jira, SharePoint, and more) to build a living map of how your business works. So rather than guess at a column like “net_rev,” an agent pulls the definition your analysts already vetted, and ranks the most authoritative when several compete (Databricks calls it “OntoRank”). The payoff shows up in their own benchmark: 84.5% of real-world questions right on the first try, versus 52.4% for the strongest general-purpose agent tested.
The second wish: Genie Spaces are growing up into Genie Agents. Customers have already built more than a million Spaces - curated, governed chat experiences scoped to a single topic. What we like most is that these agents can now carry custom tools via MCP and connectors, so they don’t just surface an insight but act on it in the systems your teams already use - and they can pull context from well beyond Unity Catalog, grounding answers in the docs, tickets, and hard-won knowledge that never made it into a governed table.
Our takeaway: Genie keeps getting more capable, which is great - but with greater power comes greater responsibility. The more tools and context you hand it, the more you need to evaluate its behavior and fix the prompts or source data behind any surprises. A confidently wrong Genie is bad; a confidently wrong Genie that’s acting on your systems and persisting misinformation is worse.
Genie ZeroOps
This is something we’ve been excited about since back when the idea was first proposed as part of “Databricks IQ”. ZeroOps is basically a first-responder agent for your production assets. When a job or a pipeline fails, the agent will wake up, analyze the error, figure out what’s wrong, test some fixes, and propose code changes inside your DevOps workflow. In addition, because ZeroOps has visibility into your Unity Catalog lineage and Genie Ontology, it can pinpoint the scope of the error, what’s affected, how critical it is, and what random upstream things could potentially have caused it – almost certainly much faster than a human spending hours navigating through all that manually. We really appreciate that it doesn’t actually apply the changes on its own; the agent is still forced to go through approvals, proper engineering practices, and running through source control and CI/CD.
Reyden
Officially the product is “Lakehouse//RT” and Reyden is the engine underneath it. Databricks wrote a new engine to power Databricks SQL from scratch, and the result is genuinely fast across the whole spectrum: tiny, pointed lookups and billion-row aggregations alike. That range matters more than it might sound, because Databricks keeps pushing past its Data Engineer and Data Scientist roots. SQL is close to a universal language inside most organizations, arguably more so than Python, so your analysts and IT admins can speak it as fluently as your hardcore developers. More people hitting the same datasets means more concurrent load and a wider mix of query shapes. Historically, Databricks SQL was (in simplified terms) mostly an interface to Spark and Photon, so it shone on big analytical queries but less so on small, real-time ones. Reyden closes that gap: you get strong performance and scalability for both the heavy analytics and the single-row, real-time hits, without having to move your data. Your engineers write one pipeline against the freshest data, and that single master copy serves every kind of workload - another place where the wall between your operational and analytical systems quietly comes down.
LTAP
LTAP might be our favorite announcement from an engineering point of view, and it's the sharpest example of that operational-meets-analytical convergence. The idea: when your OLTP Lakebase instances commit transactional data, LTAP intercepts the storage layer and converts it to a columnar format in the caching layer before it’s ever flushed to disk. Your operational systems don’t pay for it - most reads hit the cache anyway, so there’s no performance loss on the OLTP side, but that same data lands in a “big data”-ready columnar shape, immediately available to your heavy Spark workloads without a second copy floating around. The upshot: you no longer need a pipeline just to shuttle data into the Lakehouse, which frees your Data Engineers to spend their time on the problems that actually deserve it.
Where We Go From Here
If there’s a single thread running through everything above, it’s convergence: the operational and analytical halves of an organization finally sharing one governed foundation. Lakebase and Databricks Apps started that story last year by pulling the application layer into the lakehouse, and this year’s announcements - Reyden, LTAP, and a far more capable Genie - push it further, bringing operations and analytical intelligence close enough that people and agents can act on live data instead of yesterday’s copy of it.
That’s the shift we’re most excited to keep driving with our clients. Over the coming weeks we’ll be putting all of it through its paces at Origin and we’re excited to share what we learn.
.svg)