How to Govern Your AI Estate in One Place

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: How to Govern Your AI Estate in One Place · Channel: Databricks · 20:53 · Watch on YouTube

Topics: AI evaluation and benchmarks, Cybersecurity and software security, AI governance and regulation, Markets and economy, Privacy and data protection

Summary

Databricks’ Governance Hub is an account-level interface designed to give administrators and governance teams a centralized view of data, tags, costs, and AI usage. It consolidates and pre-aggregates telemetry, making it easier to investigate data access, metadata coverage, spending patterns, and AI consumption than with separate dashboards and tools. The demo shows how users can drill into cost drivers and configure per-user AI budgets, including hard usage limits. The speakers also preview planned security and performance views, deeper Genie integration, and tools to make governance more actionable.

Key points

The Governance Hub is a curated web interface in the account console, rather than a traditional dashboard imported into a workspace.
It centralizes account-level telemetry, much of it drawn from system tables and some from internal telemetry APIs.
Access is currently focused on admins: account admins can view all pages, while workspace and metastore admins have access to relevant cost and data-governance views.
The data view helps identify popular tables and views, assess their curation, and investigate user access across the account, including access through group membership.
The tags view helps teams identify metadata gaps and track tags assigned through automation, classification, and lineage propagation.
The cost view supports account-wide analysis by team, product, and workload, helping FinOps teams find contributors to spending increases.
The AI view provides observability into AI Gateway traffic, costs, tokens, requests, models, workspaces, and users.
AI Gateway budgets can set per-user thresholds and block usage after a limit is reached, with the option to configure different limits for administrators.
Planned additions include security and performance verticals, more direct Genie interaction in the account console, and recommendations and customizable criteria for enforcing standards.
The Governance Hub is in beta and can be enabled from the account console’s preview list.

Chapters

Start	Chapter	Summary
00:00	The governance visibility problem	The hosts describe how scattered telemetry and custom tools leave teams reacting to costs, access, and AI usage instead of governing them proactively.
01:36	What the Governance Hub is	The speakers explain its account-level purpose, how it differs from imported dashboards, and which administrators can currently access its pages.
05:24	Data access and tags	The demo covers popular data assets, account-wide permission visibility, and tools for monitoring tag coverage and automation.
11:16	Cost and AI observability	The presenters show how to trace spending by team, product, and workload, then review AI Gateway usage and set budgets with limits.
17:39	Roadmap and getting started	The speakers outline planned security, performance, Genie, and policy-enforcement features, then explain how to enable the beta.

Quotes

we’re trying to solve for that visibility problem. — 02:10

The governance hub is not a traditional dashboard. — 03:17

you see your privileges. — 08:35

I can actually block usage completely. — 15:02

customers can get started on it today. — 20:20

Concepts and entities

Governance Hub, Databricks, Unity Catalog, Unity AI Gateway, Genie, ABAC, FinOps, PII, Sadhana, Kristen

Full transcript (original language)

00:00 Your FinOps lead is juggling dozens of custom queries just to figure out what spike cloud spent overnight. Your governance team built a home-brew tools to track classification coverage, but those dashboards were outdated the day they went live. And your AI deployment team is flying completely blind and aware of which models developers are running or how fast they are burning through token budget. When critical telemetry is scattered across system tables, isolated workspace views, and third-party tools, your team stop

00:31 innovating and start fire-fighting. You spend half of your week playing detective, hunting down orphan tables and mystery charges, until platform admins eventually starts blocking your clothes out of the sheer self-preservation. That’s not governance. That’s a bottleneck. And that end today. And today, we’re breaking down the brand new governance hub with Sadhana and Kristen. Hi. >> Hey Youssef, thanks for having us. My name is Sadhana, I’m a product manager

01:02 here at Databricks. I actually specialize more on on cost governance and some of our AI observability. We’re excited to show you the GovHub. >> Yeah, I’m Kristen. I’m a PM on the Unity Catalog team. I’ve been here a little over 3 and 1/2 years. So, kind of making my way through Databricks. Previously focused more on our tagging system and have kind of shifted gears towards governance and AI observability. >> And I can guarantee that the governing govern tags are super critical and very

01:36 useful for when it comes to using them with ABAC. And by the way, you should really go have a look at the the I think 10,000 of features ABAC >> [snorts] >> has had for the past for the past weeks. >> Yeah, there’s some cool features out there right now, so yeah, go go check them out. >> So, tell me, what’s the governance hub, and why should users care about it? >> Yeah, so governance hub is really meant to serve as basically an account level single pane of glass for admins and then of course governance teams. Basically

02:10 right now more than ever we’ve noticed that our customers adoption of AI honestly including our own adoption of AI is just growing exponentially. And because of that the like stakes for needing visibility into what teams within your organization are doing is becoming more and more important. So where a governance hub comes in is where trying to solve for that visibility problem. So you can actually understand what are people doing within your organization, how are they using data, should they be even accessing PII, and

02:44 then also identify you know when there’s spikes in cost, AI models, and etc. maybe shouldn’t be being used. We’re really serving the like one-stop shop for observability for governance. >> Is this built exclusively on system tables? And how is it different from the existing dashboards we do have right now? >> Yeah, that’s a good question. I mean up until now you’ll be familiar with the the usage dashboard, there’s an AI gateway dashboard. These were all AI BI

03:17 dashboards that customers used to have to import into their workspace. There are a couple issues with that. I mean of course it’s customizable by the customer, but it was running on system tables on our customers on warehouse. We couldn’t really control the performance, right? It’s based on what else your warehouse is consuming, and we wanted to build a more curated experience. So the governance hub is not a traditional dashboard. Think about it like a web app. It’s a web UI we’ve curated. We’ve actually pre-aggregated all the underlying data. A lot of it is system tables. The

03:48 customers can still reproduce some of the insights if they want to go deeper, but there’s some insights uh for example on data governance we have user to group mapping that actually come from internal telemetry API support. So, we’ve just tried to centralize all of the most insightful telemetry, clearly aggregate it, and serve it directly in the governance hub. >> Uh who are the targeted user personas for the governance hub, and which permissions I am required to access it? >> Yeah, so I’ll say is this is evolving

04:18 and we’re in the earlier stages. Uh our like North Star target audience is truly governance teams, uh and we know that there’s kind of different flavors depending on what you’re looking at, whether it be cost, whether it be AI, whether it be, you know, security, and then also admins. So, for now this experience has been pretty consistently locked down to admins. Uh it does live in the account console, uh so account admins can see all pages within governance hub, but we did take

04:50 the first step in allowing workspace admins to view cost information for their workspaces that they’re responsible for, uh as well as allow metastore admins to view information related to data governance for metastores that they are responsible for. In the future, we are looking into expanding this further cuz we know it you know, when we’re focusing on admins, still narrower set, uh but because we’re introducing such wide, you know, account level monitoring, we we want to make

05:24 sure we’re conservative with who we give access to to start. >> That’s great. So, I believe we have four verticals that we’re covering. Can you tell us more about each one of them, and well, let’s start first with the data and tags. Uh what’s in there? >> Yeah, so why don’t I actually just show you? Uh I like to see something while I’m talking as well. One thing I’ll call out again, especially for people who are more familiar with working in the workspace experience,

05:55 uh this is now a new governance section in the account console. Um like I said before, workspace admins and Metastore admins can now access this experience for particular pages. So, what that means is the service surface that was previously not open to them is now available. Uh so, basically here we have four different sections. Like you mentioned before, we have data, we have tags, and then we also have AI and cost available. Um I want to highlight like two key use cases for the data page.

06:27 So, the first one is basically being able to see where your most popular tables are and whether they’re like well curated for Genie usage. So, here I can see we have two different views. You can either see kind of trend over time, how are people using assets, uh particularly tables and views within your account, and then we also have this location-based view. So, actually I can actually go in and see, hey, Fuzzer, which happens to be our internal Metastore,

06:58 happens to contain the most popular tables. And I can actually use this widget to kind of identify where are our most popular tables and help me investigate uh whether some of them are high quality, whether they’re completely unmonitored, and then also identify like, hey, maybe we should be adding tags to these tables to help describe them and help our uh Genie and other tools better serve end users. So, that’s one of our key use cases. And then the other one that I want to

07:29 hit on, which I’m particularly very excited about, is our access view. So, historically it’s been somewhat challenging to be able to see what a single user has access to, uh not only within like a given workspace or metastore, but also across the entire account because seeing group membership is a little bit challenging. So, we’ve introduced this view. It’s pre-filtered to me, and I’ve excluded some of the, you know, permissions associated with all users. But, I can now go in and see a bunch of different

08:02 securables again across the entire account, which I or this user may have access to. And I can also see how exactly they have this access. So, here we’re blending a mix of source types, so I can see I’m a particular owner, or there may be like a direct grant here. But, I can also see group membership, and that’s the key piece that historically has been super challenging. I have some exciting news that in the future we are looking into embedding whether there’s like ABAC policies in

08:35 place as well in this view, but for now it’s purely, you know, you see your privileges. >> This is very interesting. It simplifies how the people maintaining the platform or those who are granting permissions can view permission, but also understand how people are using the the data inside the platform. >> Definitely. Yeah, I talked to a customer the other day who was like, “Yeah, we just want to screenshot this and use it for our compliance review.” Cuz previously they had to go do it themselves, and it was like a whole

09:09 whole process. >> And also regarding the tags, so I think they are very important now. They can be also associated with pages, so you can add tags on the tables that are frequently accessed to tag them as, let’s say, valid. And those who are no longer accessed as deprecated to make sure they are not included in the genuine ontology. I think so, I see a lot of interesting use cases that could leverage the those the first two components. But what about I think cost and AI? I

09:43 think those are the most sensitive parts because I know we’re on the era where a lot of people are burning tokens and same thing I would apply for the cost because they’re tightly integrated. >> Before we jump into AI and cost, I do want to highlight one thing with our our tags view. Uh because tags are to your point useful and important for uh pushing what’s most important in Genie, and then also are kind of critical to representing PII in data and things like

10:13 that. So we do have this new tags pillar, and the reason why we pulled this out separately uh is cuz we found that customers were having some challenges in identifying gaps in their metadata, and then also being able to track when they’re relying on our automation tools. So we have a whole fleet of recent like tag automation tools. We have data classification, we have our two new tag automation one, and then we also are now propagating tags via lineage. So it’s more important

10:45 than ever to be able to go in and track like, “Hey, who’s assigning these tags? Can I go reach out to that person and ensure that these are the right ones to be assigned?” So just want to call that out. >> Perfect. Thank you. And you mentioned so many things, but I think we’re still missing two important parts from the governance hub, cost and AI. >> So let’s let’s switch hats for uh a second. Uh Christian was showing you primarily from the lens of a data

11:16 steward, data governance, maybe even AI engineer that’s looking to maintain integrity quality of their data. Let’s switch hats for a second and um put on a FinOps hat or any of billing admin hat. The governance hub has a dedicated cost vertical that’s been made popular among uh current customers in preview. This is a centralized account-wide view of your Databricks cross across the product lines, no matter if it’s a database, if it’s an app, or if it’s Genie. So, you come in here. This is also open to workspace admins who are generally responsible for the budgets of their own

11:50 workspaces. Um get a high-level view of of where you’re spending, um kind of what the the rate of spend change has been in the last 30 days. And I think what customers love about the cost governance hub is we both provide executive-level summary of your spend, but make it very easy to drill down into the highest contributors of cost. So, for example, say I’m looking to understand who are the highest um what are the highest spending teams on uh in this environment. I can come in, I can

12:22 group by tag. So, internally we use a tag called budget that uh represents the different teams we charge back for. Um I view that there’s actually a specific team that is two teams that are you know, materially higher in terms of spend. Now, I want to quickly identify what they’re spending on that’s stripping up cost 30% up this month. So, I quickly filter. Now, I actually want to group by product, right? Um

12:52 and I see okay, this is makes sense. This is a runtime engineering organization. They’re primarily spending on jobs. With one more click, I can actually identify uh some of the culprit jobs in the past 30 days that have uh driven up cost on this team. So, you see this one’s up 35%. Uh this one’s up 100. On average, they’re spending over $100 per day on these jobs. So, quickly as a FinOps leader, I’m able to identify what teams are spending on Databricks, what product lines are they spending on, and what are the particular workloads that I can now delegate and look for

13:26 cost optimization opportunities on. So, that’s the cost governance hub, and of course, when everyone’s thinking about costs right now, AI is top of mind. So, to that effect, we’ve built a dedicated AI AI pane. So, how is this related to Unity AI gateway, Unity gateway? Um all of the traffic that passes through Unity AI gateway, whether those are internal model services, is what we call them, or external model provider services, uh all

13:57 of that traffic, we have high granularity and a lot of observability on, and the governance hub provides an executive uh scorecard report on all of your AI traffic and consumption. So, all of this is built off the same governance traffic and observability that um you also have available in system tables for the AI gateway. I come in here, I’m able to quickly identify how much have I spent uh in the last 30 days on on AI, what are my potentially they’re internal

14:29 models as well, and uh who are the users that are driving up AI costs. So, I’m able to get a similar time series view, group it by model, model family, user workspaces, um and look at spend, but I can also just look at token and request count as well. Now, observability’s great, but, you know, obviously for AI governance, that’s not enough, because AI costs can go exponentially. So, we’ve also built um budgets for Unity AI gateway. This is available in the governance hub for account admins and workspace admins as

15:02 well. Um so, with these budgets, you’re actually able to come in and create a budget that’s targeted at Unity AI gateway. You can even specify if you want Genie specifically, because we’re looking to integrate uh Genie better in the gateway. Uh so, in this case, I’ve spent uh I’ve created a budget covering all my workspaces targeting Genie spend, and able to come in here and actually set per user uh thresholds. And it’s not just a soft alert, I can actually block usage completely. Say every user should only get $1,000 of

15:35 Genie. Um I can add that there. I can block them after that. But um maybe I want to give myself a higher budget because I happen to be an administrator as well. So, I can give myself $2,000 a budget. There we go. That’s a hap- that’s a setup I’m pretty happy with. So, what what’s the output of that? Uh when you create a budget, um and actually starts tracking your usage, you’ll actually get um a I actually almost said a near real-time frequent update on

16:06 your users, what they’re spending, what what they’re spending against their threshold. So, here you’re sort of seeing we set a we set a $10 limit. And um our internal users love Genie, so we’ve already surpassed that uh limit in the first couple days of the month. So, I’ll actually come in here and increase the override for some of these users. So, that’s kind of going back to AI governance, it’s dedicated observability, hard caps of budgets, and a lot more insights, and um

16:36 verticals coming. >> Tracking the cost is now easier than ever. And I really like the fact that we can drill down going from the higher level cost through the like the most granular um parts where we can where we will be able to understand from where the cost is coming. Whether by workspace, by workflow, by by product. And I also think about when we move to the AI part, with uh the budgets, we will be able to

17:07 avoid any accidental spend by granting, let’s say, a budget to every single user, and of course overriding for the super user to make sure that they are still able to achieve uh the task that has been given to them using using AI. And I really really like I so you should definitely use the governance hub and I’m really keen uh to uh to have an overview about what’s coming soon, what do you expect to add for on the on the road map? And

17:39 also Genie is now everywhere. How do you intend to integrate Genie uh uh in the governance hub? >> There’s a couple of different ways that we’re expanding governance hub. The first being expanding to other verticals. Uh so basically we have two key priorities that we’re rolling out this quarter. The first one is security. So some of the folks that listen to your podcast may be familiar with our security analysis tool that’s uh I think some folks from field

18:10 actually built. And we found a lot of success with that feature uh and we’re actually going to bring in a lot of the metrics and stats that we’re tracking against it into the product and that will live in governance hub as well. And then there is another one where we just call out. We actually are introducing a private preview for this which is a performance vertical. So if anyone’s particularly keen on kind of understanding performance opportunities, uh looking at like specific warehouses, uh we are doing a small

18:44 um preview before we introduce that to the the main governance hub experience. And then there’s two other areas that we’re investing in. Um I’d say one you mentioned which is the Genie integration. So today for users in our experience we do have a Genie one integration. So you can actually click and ask Genie button directly in the UI and it will take you to the Genie one surface where you can ask questions and they’ll be powered by our governance hub specific

19:17 skills. However, we did hear a lot of feedback around people wanting to interact and ask questions directly in the governance hub UI. So, we’re working with our account console team to introduce a Genie code integration directly in the account console. Um so, more on that coming soon. I believe the latest target date was early September, so uh should be available shortly. And then the final one, uh which I’m sure Satyanarayan can also share some thoughts on, is uh we want to make this experience more

19:50 actionable. We’ve been very focused on making sure uh account teams and uh governance teams have visibility into what people are doing, but now we want to make sure that you can ensure your policies and standards are enforced. So, we’ll be adding more DataBricks DataBricks best practice recommendations, as well as the ability to kind of define your own criteria. >> This is really interesting. Can’t wait to see all all those innovations popping up on the governance hub. If I’m a

20:20 customer or a developer, what are the prerequisites to access the uh govern or to get started with the governance hub? >> Yeah, so uh governance hub, as of a couple weeks ago, is now in beta. Um the team is moving fast, working hard to make this generally available as well. Um it’s an account-level preview. So, if you in in the account console, if you go to your preview list, look up governance hub, you can enable it from one click there. Uh there’s sort of terms of services associated um in that preview around data centralization you can look at, but customers can get started on it today.

20:53 We’ve already had great feedback. We have lots of customers using it. Governance is no longer an afterthought on DataBricks. It’s a first-class product, so we’re really excited to hear your feedback on governance hub and appreciate your time um learning more about it. >> And if you’re watching this video maybe in 3 months or 4, the governance hub would maybe be GA and you’ll have it by default on your account console. Thank you so much Sadana and uh, Kristen and see you soon.