# Cloudera's AI-Driven Data Strategy and Career Shifts

**Podcast:** Chain of Thought | AI Agents, Infrastructure & Engineering
**Published:** 2026-11-08

## Transcript

I knew I was going to ask my entire team to automate their jobs.
I knew how popular that would be.
I knew just how great I would be viewed by my team.
They were going to love me.
The more we automate data analytics, the more valuable we're going to be to the business, ironically.
So I came up with three career paths.
One is AI engineer.
I got a guy named Laszlo in Budapest.
He's done this.
He used to be just a data architect.
He says just a data architect.
And now he's a full stack engineer.
He couldn't be happier.
Automate your last job.
What are the things that you did?
Generate synthetic fake data.
Guide AI through the actions of what that job was.
You can give everybody in a company a dashboard and still leave them waiting for someone to make a decision.
Adding AI to that dashboard raises a harder question.
When do you trust it enough to take the next step for you?
Welcome to Chain of Thought.
I am Conor Bronson, your host.
Joining me today is Shade Christian, Chief Data and Analytics Officer at Cloudera, where his team is building the systems that we'll be talking through today.
You've described customer support information that can tell someone they need training and professional services.
Are there particular areas where you're seeing the most customer signal or interest?
in this new analytics era?
Yeah, it's an interesting question.
I think that the most actionable signals can be buried very deep in the data.
I can give you an illustration of one that we uncovered and how it's working for us.
I think your audience should first understand kind of what we do, Cloudera sells data platforms.
We have a very complex 60 micro service.
a full service data platform and it's complex because our customers who are the most recognizable brands and largest companies are trying to solve the world's most complex data problems with them and because of that we have a big support group an organization that supports them And so we take a lot of case data from our customers and diagnostic bundles and things of that nature.
And so typically our professional service and training team would shadow, as is usual, the GTM team, the go-to-market team.
And they will go out together and determine whether there are professional service or training needs for our customers to help them with these complex products.
But the truth is the secret to who really needs professional services and training for our customers is buried in the case data.
If a customer opens three cases per quarter on NIFI, for instance, one of our components, and we solve them as non-break fix NBF, it means that there's nothing wrong with the software.
We had to help the customer understand how to use it with their other systems and components.
And so if you look at these signals, We can then provide and do in a single pane of glass called shared customer intelligence.
We provide this information, these signals about which customers are struggling with which products and may need professional services and training.
And we offer that in a single pane of glass for humans in the loop to react to.
Before we dive in, I want to say a quick thank you to our presenting sponsors for Chain of Thought.
Ingest for long running agents and Walrus Memory providing portable agent memory.
Plus, this episode's other sponsors, Coralbrex.ai, providing lightning-fast open-model inference, and Wisdom AI, the enterprise agent analytics platform.
So, great.
We've got a single pane of glass for humans to look at.
How are agents leveraging that data on the back end?
And how are you enabling not just human coworkers, but the growing group of digital coworkers that we have built around them?
Well, first, we're utilizing multiple layers, our AI intelligence layers actually communicating with the cases and actually crunching the data, getting the sentiment.
So there's a lot of agentic AI working to understand and produce the signal.
The single pane of glass, it becomes then a human affair.
What we do have the capability is going beyond that.
In fact, we've programmed agents, Connor, to...
create an opportunity in our CRM so that the CRM then alerts our PS&T organization that they have an opportunity for this customer utilizing this technology.
We also have the capability to communicate with our customers via that CRM agentically, which says, hey, here's a...
a coupon for 30% off of NiFi training, and we can automatically send that to the customer.
The challenge for us is that like the regulated customers that we have, we're pretty conservative.
And so the business owners who own these applications are still looking for a lot of assurity in terms of reliability and capability and accuracy before they're allowing our AI agents to be released into production.
Interesting.
Can you unpack a bit more what kind of signals your enterprise customers who have these concerns are looking for to enable new AI capabilities?
What are the blockers holding them back?
It's the data privacy and protections that our culture at Cloudera has because we want to mimic.
the highly regulated environments that our customers are held within.
And so just by our nature, we're very conservative.
Number two, well, you have to make sure that these AI agents are getting it right.
Are you sending the coupon to the right customer?
Are you opening the correct opportunities?
Is it in the correct amount?
Does it really describe what really needs to happen?
Is it clear for the professional services organization to understand what's happening?
We have a really good track record currently of passing those tests, but we're still kind of in that cultural feel it out.
What are we going to allow agents to do in production?
I want to talk more about the data architecture behind this, because I know there are crucial decisions being made around, you know, what goes to a non-deterministic system versus a deterministic workflow.
But.
I'd love to understand a bit more first about this shared customer intelligence view you've created and what was happening to customer signals, whether that's case notes, account history, anything else you're pulling in beforehand and then how you're now showcasing it today.
Right.
So these signals were generally not available, right?
Because you don't have the professional services group or the go-to-market community or the marketing team ferreting through case volume, right?
It's just typically unavailable.
And so we work the same way we've always worked.
We get these ideas from our business customers.
Hey, can't we look at case data, et cetera, et cetera.
And so we go and stick our technology on it, particularly generative AI in this case.
And so previously, these signals weren't being provided.
Now, in terms of the signaling, the way that we use shared customer intelligence as a single pane of glass, I mean, have you been to the Bellagio?
Have you seen the fountain where it creates shapes and then the shapes disappear?
This is kind of our image about what we're trying to do.
My personal philosophy is that...
I created a great career based on dashboarding and data and analytics, right?
We trust them.
We love them.
The great part about them when they started and through their evolution is that they pulled together data and insights from places you couldn't see together.
And so they've been a marvelous evolution of what I call in decision intelligence.
But even when you look at dashboards, my current feeling, Connor, is that they're just another system.
They're just one piece of data, one piece of the puzzle that needs to be put together for, for instance, professional services or a go-to-market account team to make a decision.
And so we're trying to get away from dashboarding and data and SQL and Python coding as a way to find the puzzle pieces.
And we're trying to surface intelligence with dashboarding to me just being another system.
And so here's where the fountain idea fits in, right?
This single pane of glass, you might say, well, it's just another dashboard.
It's just another system.
What are you talking about?
Yes, but it's intended to be the Bellagio fountain.
It's intended to surface these signals temporarily, provide them to our business team for them to determine what agents can accomplish or what automation can accomplish or where in the workflow this automation can happen.
And as soon as they make those decisions and we...
automate those tasks, then the fountain comes down, the shape comes down and there's nothing anymore.
And there's just actions being taken on behalf of the business.
So I think this is a really important point around how we need to enable business users within an enterprise and the types of data that we need to pull out for them.
And the fact that a lot of that has been hidden at times in the past or hard to access, you know, I would say be like you mentioned specific tools here like uh having to learn a lot of SQL etc like frankly not to say you it's not useful to know SQL today but a coding agent can write your SQL for you if there's something you need a query for you can describe it in plain language it can go execute that for you so it's it's enabled many business users who were less technical to take on data tasks that they wouldn't otherwise but I think there's still a big opportunity for engineers here and a lot of my audience are developers they're high technical But they may or may not be thinking about how can I enable the business side of our organization?
What would you say to engineers who are hearing this and saying, OK, that's great.
There's more data being exposed.
We want to enable people at the org.
But it's not really necessarily a priority for me.
There's got to be an incentive, I would assume, for them to help support these kind of systems and build off of this data because of what it could enable for them in the business and their own careers.
So what is in it for your engineers is, first of all, that it seems threatening, but it's not threatening, that the business can now self-serve on, for instance, dashboarding, right?
Dashboarding can now be accomplished.
My team has accomplished it through natural language.
And if we really want to be valuable to the business, you could do two things at Cloudera.
You could speak.
SQL or Python, you could learn the data, you could learn a visualization tool, and you're still going to spend hours and weeks to create something valuable for you to look at.
Or you can't do that in your natural language, but you have to wait in my team's queue for weeks and months.
We have a tool that enables people to create their own dashboards just using natural language, and it takes them minutes and hours.
So the first thing for the engineers in my group was...
Not that this was the cause and effect, it was really in the reverse.
But let's get rid of the problem solving and visualizations.
Let's get rid of the I'm constantly perpetuating new visualizations for the business and enable them to do that so I can get on to higher value work, which is AI engineering.
As folks are enabled to spend more time at a different abstraction layer and kind of push off this work that has previously taken up a lot of hours, we're seeing very interesting opportunities to put teams on work that is higher value for their skill set.
And I know one of the areas that you have seen advances through your work has been around enabling faster onboarding and the ability to take on more accounts simply because of what technology is enabling for Clutter.
Absolutely.
So one of our business customers is this support organization that I talk about.
I talked it from a PS&T perspective before, but we support that support organization.
And we also support the customer service managers.
And the challenge with our customer service managers is that our, again, tools are very complex.
Our customers have very, very complex environments.
And so it would normally take a week to get up to speed on a new account.
And they can only take on two to four accounts depending on how complex the accounts were.
That's how intense the customer service is in our neck of the woods.
And so what we've done is we've used AI to look at the sentiment on all of our case tickets, summarization of all of our cases.
And now it takes only an hour or two to get up to speed, which means...
That's 95% savings and effort and time.
And now our CSMs can support up to eight or nine customers at a time.
That's 267% improvement.
I mean, that's real tangible value.
And not only that, Connor, but we've enabled the top five knowledge base articles and the top five past solved cases.
And now the mean time to repair for tickets is decreasing because we're giving them the intelligence they truly need to solve the case.
And a net promoter score is increasing with that customer community as well.
So this is really interesting.
You have, through these technological adaptions, improved the ability for multiple parts of the organization to take action.
You know, we haven't touched on a few parts of the org.
I mean, obviously, things like recruiting, hiring, et cetera, are very important as well.
Ops.
What are you seeing for folks who are?
non-technical business user side of things, outside of even like the data analytics team, are they seeing major gains from the internal systems you've built?
Some, some are seeing major gains.
I think that there's three levels of business value that you can really drive with AI, depending on your level of commitment, your tech feasibility.
and your risk tolerance.
One is these individual productivity efficiency gains.
You're wiring up a foundational model.
You're allowing people to affect their jobs, which is great.
We've seen a lot of productivity efficiency there.
Those are table stakes.
You're not going to change the game necessarily with those use cases.
The second level is what I'm referring to and guiding you through in my illustrations, which is workflow automation.
It's not just one person's productivity.
It's looking at how do CSMs operate?
What is the decision intelligence they need?
Let's automate their workflow and give them all the puzzle pieces at the moment they need decision.
And then the third type of use case is really a game-changing use case.
And one that we have that's very effective is this automated data dashboarding capability.
We call it natural language querying tool.
And with that, we are seeing, again, minutes and hours of dashboarding capability.
It has revolutionized the...
that my team has to develop these AI applications.
And so, yes, sometimes massive, sometimes very effective and very productive over time.
And some initial productivity efficiency gains are the immediate business impacts of AI.
Season four of Chain of Thought is presented by Walrus Memory, the portable memory layer for AI agents.
Your agent has learned your code base and how you work.
Now you want to use that context in another tool.
Exporting a file gives you a snapshot, but what happens when the context changes?
Walrus Memory is a portable memory layer that lets your agents store context for later use across tools.
Python and TypeScript SDKs plus native MCP support let you connect it to your agents no matter where they are.
You said who can read and write, so sharing context doesn't mean opening up your whole memory store.
If you are building across tools or model families, take a look at Walrus Memory.
Learn more at walrus.xyz slash C-O-T.
That's walrus.xyz slash C-O-T.
And a lot of what enables these gains is getting the underlying information architecture correct.
Because if you're feeding the wrong context and data into AI tools, it's simply going to go off in the wrong direction.
As smart as frontier models can be and as smart as we can build these systems.
they're still reliant just like a human is on what's the information that I am being provided.
And I know that there are efforts that you've taken to ensure that there's extreme trustworthiness behind not just the self-service aspect of what you're developing, but also understanding the limits of any accuracy and privacy claims.
Can you walk me through what happens when, say, an employee types?
question in and is looking for an answer like what's happening with the model where is information coming from uh what stays outside of it you've backed into a very strong problem statement right i think that most everybody in my role and in the units that i um Manage came up with the idea years ago.
Well, let's put this GPT-like interface over our enterprise repository, make it searchable, and bam, natural language speakers are able to generate insights instantaneously.
I think we've all had that ambition.
I know of customers and a competitor who did it what seemed to be the logical way.
Well, let's put that chat GPT-like interface on top of the enterprise repository.
Here's the problem.
It knows no context of your business data.
You're asking it to look at a wide swath, a broad data estate and come up with answers.
And it's not capable of doing those things.
It hallucinates wildly.
And customers and competitors I've read have put...
dozens and hundreds of data engineers to try to figure out these solution nations and make it work.
We used a very different approach.
I assigned seven data engineers to this problem.
And, um, We said to ourselves, how do we achieve this data privacy?
I cannot allow a public model to look at my private data.
We're self-regulated in that way.
That really makes it difficult for me to get a GPT-like interface over our enterprise repository.
So data protections and privacy is critically important, but so is accuracy and reliability.
We've taken a look at what people have done in the industry and how wildly inaccurate it is.
And so what we've done is we've created a solution that's barely AI.
How did we come up with it?
We were backed into a corner.
How do we do this?
And we sat around and said, wait a minute, we've been producing SQL and Python code for the last dozen years that comes up with the exact answers to these questions.
Is this really an AI problem?
Right?
How do we achieve data privacy?
How do we give...
total accuracy and reliability in the responses because our senior executives and all the way down to individual contributors are making key decisions for the company based on this data.
We want it to be accurate higher than the accuracy and reliability rates even at the great foundational models today.
So what we've done is we've created the interface.
Connor says, I want to know about ARR, which is our golden metric, which is annualized recurring revenue, right?
And we do use a foundational public model to vectorize the question.
That's it.
Translate the question into a vector array.
We have that model.
Compare it to business and technical metadata that we have over time heavily involved.
And this really gets to the answer that you're asking about, Connor, because It's not good enough that I have data in a table and I understand the schema and I have great business glossary and data catalog information.
I can tell you how to compute ARR.
I can tell you where to go and what table, how to compute it, what it means to the business, et cetera.
But AI still doesn't speak that language as much as it speaks natural language.
I'm talking about generative AI.
And so we've really worked on that business and technical metadata.
And the AI says, okay, I understand the question that's being asked.
And I understand.
Where on premise in the private environment with shades, AI agents that are behind the wall, air gapped, what they need to do to call the right SQL and Python code to get the correct answer.
Those agents take the instruction.
They generate the answer from the SQL and Python.
They give it to the user around AI because AI is not allowed to see the data.
They give it in text tab.
They give you a, um, visualization which you can then pin to a dashboard that's how the dynamic dashboarding and automation of dashboards comes in they give you the sequel they give you the lineage and they say have a nice day interesting so you're using the frontier model to vectorize um it's querying i presume a reg database to pull this information out and then you have a very intriguing approach to privatization here to ensure that data is not being exposed and i know that's been a huge concern for a lot of other organizations as they've tried to leverage frontier models i mean we've seen examples like let alone like small things there was the conversation about grok essentially stealing customers code uh to use for training data a couple months ago and it's not the only example we've seen of data leaking by any means but there are major privacy concerns when it comes to enabling AI at a major enterprise.
I'm curious if you have advice to other leaders who are listening and thinking about improving their own approaches to private AI and leveraging it within their tech stacks.
I do.
I want to say where my advice comes from because it's heavily informed, right?
So because my company develops data platforms, data analytics and AI solutions, and that's what we sell.
And because my team is the data analytics and AI team, we are customer number one.
We utilize our products and we are to be the best in reference example for what our customers should follow.
And so half of my job is beating the street.
talking to prospects, customers, helping people and supposedly advising them on how to do this.
But the truth is that I deal with some of the most advanced companies in the world and they teach me as much or more than I teach them.
So this is the advice coming from the top telcos in the world, the top banks and the top healthcare institutions, financial services, the largest companies in the world are who our customers are.
And what we're all doing together is we're privatizing AI, which is two things that you saw in the NLQ.
It is taking AI, the foundational models, a lighter version of it, embedding it into our on-premise air-gapped systems so that we can share data with those models because we're not going to share them with the public models.
So where you AI is critically important.
These things now with the Cloudera AI inferencing engine are getting very capable from a performance perspective.
And so we're able to allow AI to see an enormity of our data estate from a compute and storage perspective.
The second thing you ask yourself, my advice based on our customers is.
If you AI, not just where you AI, but if you AI with NLQ, what it's barely an AI solution.
I have 200 power users, Connor, on this application only, but my team uses it day in and day out as their primary development tool.
And talk about winning the token war.
This application with that user community costs us $2,000 a month.
That's it.
It's barely AI, right?
You're building on the foundations of everything you've done before.
You're allowing the accuracy and reliability to be in the code base that you've already established for decades.
And you're using AI where it makes the most sense in a private environment.
And you're discerning, high discernment about if you're even employing it.
Because as you mentioned at the top of this call, you raised a question in your head about what's deterministic, really?
What's predictive?
That's an important part.
of if you ai this brings to mind a couple questions for me one uh when you were building the system i'm curious did you use ai enabled uh engineering to do this or as you were coding parts of it where was your team using an internal or external tool for the coding side yeah we do we've used a bevy of them you know github copilot we're we're heavy into cloud code right now my bi team uses cursor quite a bit But again, it doesn't have access to the data, largely speaking, right?
So we have these data privatization masking techniques that we use to get the assistance from the coding agents and become much more effective and efficient coders while keeping our data private.
And it has achieved, in some cases, 75% or greater efficiency in coding.
We absolutely use those tools.
Yeah, I think that's...
very fascinating to see how rapidly AI has taken over coding in particular, because it's somewhat solved domain as far as, hey, we have a clear win condition, I suppose you could call it, of what we're trying to do with this code.
And we know what good looks like.
We have so many examples of it.
And it seems like an obvious area to continue to invest.
But to your point, you know, there are still privacy concerns that come with that, as I think is a key theme for this whole conversation.
Beyond that, another, I'm sure, challenge that you faced around this was ensuring that the glossary and metadata information that your systems are pulling from for these answers is accurate, is up to date.
How are you managing context?
Because I know for a lot of teams, that context and data management can be a challenge that will unwind the value of these systems.
Even if they do incredible AI engineering to backstop the system, if the data inputs that are going in are wrong, they're still a problem.
It's funny, I advise our marketing and sales units when I'm not at a conference, I'm going to be talking to someone, what can I talk about, your peers?
And I go into a little bit, but I always say there's always two forever topics with CDOs.
Like, what is your data strategy?
Everybody wants to know everybody's data strategy as if that's, we can even define it.
And the second thing is data governance.
Excuse my French, but no matter how long we've been at data governance, we all still suck at it.
How can we put so much enormous resources and fight the systems that be for budget and concentrate on this so much and we still suck at it?
The good news for AI.
is that what you've done over the last decade or two for data governance really applies to AI governance.
What you've done for data readiness really helps you get AI data ready.
That's the really good news.
The bad news is that AI doesn't really understand naturally the ways in which we've described and the way in which we've governed data.
Yes, our business glossaries are plain English.
Yes, we've got easy to understand technical calculations, but it really needs a lot more help because of the amount of data that can understand and reason with it at a particular time.
So what we've done on top of it, Connor, is we've provided a lot of different naming conventions.
We renamed our tables, like, even though it's P underscore finance, you should understand it's the finance table.
Even though the column is ARR, it should know it's really annual recurring revenue and it should know it's revenue and it's point in time revenue.
It needs to understand a lot more detail.
So that's that.
We also included different columns for synonyms.
We've also physicalized many data sets.
Whereas you get a lot of benefit from your data governance and data readiness.
programs, you have to take that next step to make AI understand the business meaning, not just where to go, what to calculate, the business meaning, because it's all about that prompt that's coming and how to equate and semantically match what's going on in the technology with what the user wants is a struggle.
This episode is also brought to you by Choralprex.
Coralbricks serves open models through one API at the lowest prices and the fastest speeds.
DeepSeq v4.1 Flash streams at over 300 tokens a second.
Plus, Coralbricks also just put out MotionBench, a benchmark that asks a simple question.
Can a model animate?
I've personally contributed to the open source MotionBench and was impressed that DeepSeq v4.1 Flash managed a close third place.
behind a leading Claude Opus 5.5 at about an eighth the cost per cask.
The animation for this ad came out of that benchmark.
We used Claude Opus 5.5 and DeepSeq V4.1 Flash via Coral Bricks to create and edit it.
Get lightning-fast open model inference at chainofthought.show slash coralbricks with a free $5 in credits on signup, plus an additional $5 credit on first purchase.
How are you evaluating the success of the system?
Do you have regular evals you're running?
What were you doing for testing, etc.?
Yes, I mentioned before that even though I think we're knocking out of the park with shared customer intelligence, I'm still unauthorized to take these actions with agents in production.
And that we measure just how accurate it is.
We look at the logs.
We see what it didn't do right if it didn't and why it didn't do it right.
We correct and modify the solution.
With NOQ, it's very easy.
You've got a thumbs up or thumbs down.
And because it gives you the SQL and the lineage, we can see exactly where it's gone wrong if it creates a SQL or an amendment to it.
what I told you before.
Yes, the agent, the public model is vectorizing and sending instructions to an AI agent, but when it can't find a match, it's authorized to generate its own SQL and give it a try.
And then we disclaim when that answer comes that this is AI answering the question and not CDAP, our enterprise repository.
So we have all this data that we're getting back.
human feedback as well as the logs we keep and the accuracy and reliability we're able to make changes based on that feedback.
And as you think about this next stage, Shade, where you are going to begin enabling more agentic actions potentially on top of this data or based off it, what are the first steps you expect to enable and what are the...
I guess, blockers that are still in place there.
Is it simply a, hey, we're at 94% accuracy.
We need to get to 99.9.
Is it something to do with the privacy side?
What's kind of the next step you're building towards?
Well.
this threshold calibration is very touchy.
What it means is that the controls you can place on public models and embedded models, especially the foundational models that are extremely beneficial, are very sensitive.
And so do you say, all right, if I'm 99% to the prompt or if I'm 94% to the prompt, these threshold calibrations are very touchy.
And so what we're learning is that we have to even do a better job of creating a semantic layer.
The technical and business metadata was very important.
We made a lot of strides in that area.
It's not like we tried to get out of creating a true semantic layer.
We're building toward it.
But you have to continue to evolve just like data governance, data readiness.
It's hard work that you have to keep going through.
So the immediate...
Next steps for us are accuracy and reliability in the hands of the AI engineers based on the feedback from the users.
In the long term though, Connor, what we're looking at is pretty advanced concepts around enterprises code and the business self-serving, not just on dashboarding, but on business workflow automation.
And that's terribly exciting for us.
We just signed a deal with Mistral.
And Mistral, Cloudera, and a huge SaaS provider are now currently working on a way for Cloudera to bring the data context through Mistral Advanced Reasoning and take agentic action in multiple SaaS systems across your enterprise.
And so where we're headed is essentially allowing the business to use natural language in their CRM or ERP systems to say, I don't think this workflow, it's requiring us to do too much analysis, requiring too much data entry.
Work with me on proposing where our business workflows can be automated to save us very valuable time.
I think in the US, at least, we hear a lot about OpenAI.
We hear a lot about Anthropic and, you know, we get Meta and SpaceX a bit.
There's a lot less conversation about Mistral and the value they're providing.
Can you tell me more about how Cloudera and Mistral are going to be working together?
Yes, you hear less about OpenAI and less about Anthropic in Europe, in the EMEA region, Europe and the Middle East.
Mistral reigns.
They are much more capable of working with the regulations and the impositions imposed by regulators and data sovereignty.
They've got these things solved more effectively.
And so with our European and Middle Eastern customers, they're pretty steeped in Mistral and more mistrusting of the foundational models that come out of America or China.
And so we're, we now have a huge data sovereignty play, which it starts with Infra, which is why we're with NVIDIA and Della's partners.
It continues with our data platform and then it goes to AI sovereignty, which is part NVIDIA, part Mistral and part Cloudera.
And so we're trying to make that stack data sovereign.
So they have the advantage in those capabilities, Mistral.
Very interesting.
As you think about the next, stage of development here.
Are there particular capabilities you're looking to build or to partner with Frontier Labs on to continue to extend the approach that you're taking today?
Yes, I mentioned that POC that we're doing.
I think that would be revolutionary in what this concept that we're trying to prove with our partners is great.
You have CRMs and you have dot AI companies who are giving you the power to leverage AI in a slice, a vertical of the business.
And even the largest SaaS providers only have 20% or 10% of the data estate available to those agents.
Cloudera has 30 exabytes under management.
That's more than any other hyperscaler.
We want to work with those SaaS companies and Mistral, and we want to leverage, and what we're proving out based on that case solver tool that I mentioned earlier, we're proving that SaaS companies with their agentic capabilities are significantly more effective if they can leverage the data estate and the AI capabilities that my team's developing.
And so that's...
Enormously powerful if SaaS companies are using foundational models and the entire data estate to take the right actions across the business.
Season four of Chain of Thought is presented by Ingest.
Long-running agents need endurance.
They must endure waiting for users and weather tool failures.
They must take each new jump in model intelligence in stride.
Ingest makes long-running agents durable, observable.
and improvable over time.
Check out their generous free tier and a sleek local dev server for testing your workflows.
Grab the ingest link in the description or visit ingest.link slash cot dash pod to get started.
Build for the long run with ingest.
Any frontier model can query your data, but can you trust its answers?
This episode's sponsors Wisdom AI put that to the test in their latest benchmark study.
Alone, Claude Code topped out at 73% accuracy.
Wisdom AI closed the gap, delivering accurate answers 3x faster at a third of the cost.
I sat down with Kapil Chhabra, Wisdom AI's co-founder and CPO for episode 75, and he explained the analytics architecture that makes these kind of results possible.
Check it out for a full deep dive.
Plus, get the blueprint for trusted AI analytics at wisdom.ai slash benchmark so you can deliver accurate answers on Claude, Chet2PT, or wherever your business gets their insights.
Download your copy at wisdom.ai slash benchmark.
So as you think about these new leverage points, it certainly is changing the roles and the responsibilities of different team members.
As we're able to extend ourselves in certain domains with AI, it means that, as we mentioned earlier, the level of abstraction you may be working at may be changing, or maybe your focus areas for roles changing.
How are you adjusting the responsibilities of your team or your partners to work with your data systems and ensure that you're continuing to deliver not just analytics value, but you're moving towards these next stages?
How am I doing it clumsily, I can say?
What happened to me was I was not a late adopter of generative AI, but I was not on the hype train for the first six months.
And then, oh, wow, like this is real.
When you start experimenting with it and you understand it's real, my processor got overloaded seeing the future of data analytics being automated.
And I kept it to myself.
I talked to my CIO about it.
I kept it to myself and started really thinking about it.
And I think this is a very important distinction for any data leaders or tech leaders listening.
When you realize that this technology is going to change everything you and your team do, it is a leadership decision and a leadership commitment to understand it, to plan around it, and to shift your strategy and the safety of your people toward it.
And so we started working with it.
I knew I was going to ask my entire team to automate their jobs.
I knew how popular that would be.
I knew just how great I would be viewed by my team.
They were going to love me.
But before I did it, I came up with the three career paths that I was going to pitch them because I did realize that the more we automate data analytics, the more valuable we're going to be to the business, ironically.
So I came up with three career paths.
One is AI engineer.
I got a guy named Laszlo in Budapest.
He's done this.
He used to be just a data architect.
He says, just a data architect.
And now he's a full stack AI engineer.
He couldn't be happier.
I invented the second career path of parachutist.
I had a data engineer in Costa Rica who was adept.
And he said, you know what?
I don't want a vibe code in front of customers.
So I want to keep my mind firing in a Python library all day.
Great.
Where do you want to work?
How can we get you there?
What can we do?
And we actually helped him off board into someplace that he wanted to go.
And we weren't happy to see him go, but we were happy to be part of his decision and help him through his decision.
And the third, in Costa Rica, I have an amazing data analyst who is really understanding the business.
And data analyst Durano Q is really automated in a high...
percentage of the job is automated, but we've put her in the business.
She sits with in the forecast calls in the QBR.
This is a little hypothetical.
I'm exaggerating a little bit, but in the QBR, she can sit there with her multiplicity of machine learning algorithms and she can listen to the CRO talk about where he's going to go and what accounts they're going to go in.
And she can raise her hand and say, you know, My propensity to buy model doesn't show that.
By the way, this company went with a competitor six weeks ago.
And I think these are the accounts you should go to.
And in that moment, she becomes an insight assassin, striking at the very moment when the true analysis is needed with true insights that are available to her through the Bellagio fountain in certain panes of glass.
And on that one day, should she be able to do that?
She would justify her salary for a year, which I could have harvested if I wanted to lay people off, right?
So it's a leadership act to understand where you're going, create these career paths, and get your people the knowledge and understanding they need and the talent and skill to get where they need to go because my team are far more valuable to the business now than when they started.
And I can imagine, though, that...
The transformation process is challenging for some folks.
Some people are very excited by it.
Some people are resistant.
How are you managing that change process with your team in particular to make sure that they feel comfortable in this transition?
And how are you enabling them to feel excited about it?
Well, I think that's what I meant with clumsily.
It's a leadership act to plan for your people.
And I didn't promise anyone I was never going to lay anybody off or that the world wasn't going to change for everybody.
But what I did say, and I've expressed my belief many times in their capability and how they are going to be more effective.
Look at natural language querying, right?
It automates my data analyst role, and now the business can do it.
So why would you need the data analyst?
because they know the data better than anybody.
They know where to find it.
They're better with AI than the business.
And if they get their proximity to the business and understand it as an insight assassin, they're going to do better for the business than the business can do for themselves, period, right?
So I was very open and honest.
I was very transparent about where I thought I was going.
I was very acknowledging how unfair it was that I was asking them to...
automate their roles.
And to be honest, I have one team who still doesn't believe me.
And this is two and a half years later.
They're all still here though and doing well.
And so there's a campaign you have to have.
There's a trust that you and your people have to have together.
And it's not always there.
And it's been difficult, Connor.
It's been one of the most difficult transitions.
I go back to thinking about how human value has been changed.
based on AI.
And I have these cool graphics in a speech that I do, and it shows that roughly 60%, 65% of what humans do today or did yesterday for human value is, well, we execute tasks.
We analyze, we coordinate, we adhere to processes, which means more data entry.
We do research, we generate content, and we perform coding tasks, right?
These are all very important human values, or so it seems.
But AI is changing all of that.
Those 65% will decrease significantly over time.
And the human values that will be more important in the age of AI are empathy.
and judgment, and insight synthesis.
I give you all these fountains.
You have to synthesize these insights, and now you're deciding more than you are thinking and providing data to decide.
System design thinking, therefore, becomes very important, as does working with AI.
That's model monitoring, AI orchestration, and AI governance.
And so yesterday's human values are changing to today's human values.
And when we think about how people adjust, yes, it involves adjusting to the role, but it also involves really adjusting within yourself what qualities and attributes you have that are going to come to the fore.
This is a really poignant question because I think it's easy for us all to get caught up in elements of this execution that is so much of our work and fail to...
spend the time learning and growing that we need to, or at least not be as intentional as we perhaps should be.
How would you recommend that folks think through this expanded responsibilities and capacity that we have with these new tools that are enabling us?
I don't think it's difficult.
I think just like I keep saying how clumsy I've been at change management, change management is so important and the primary adjustment to be made is to mindset.
And so the first thing is to adjust your mindset.
And it's not that hard.
You'll notice that I mentioned as yesterday's human values, process adherence, coding.
So wait, speaking machine is a human value?
How did that ever become a human value?
That's a machine value.
And so if you really look at this, There's opportunity to get rid of tasks you never really wanted or tasks that are unnatural to us that we've had to learn anyway.
So the first thing is really about changing your mindset.
The second thing to realize is that, well, do we really need to examine and improve these skills and attributes within ourselves?
Some of us may be more empathetic than others, but that's a core human capability.
Judgment.
Decision ownership?
We already have that.
This isn't just about coders.
It's about the business too.
Does the business make deals?
Do they have to know where to spend their time?
Yes.
They're just doing more of it day to day than they used to.
So do we need to give them an insight, synthesis, class, decision, ownership, workshops?
No, we don't.
They already have judgment today.
They're just going to be relying on it more.
Sure.
Even when we as techs and when the business gets involved in AI monitoring, AI orchestration, AI governance, there is a learning curve to be done there.
But really, you've already struggled to make your systems behave today.
It's a behavior that you already know.
And we're just going to give you differences in how this technology behaves.
So I think there's a monumental shift in terms of mindset.
But once that mindset shift is made.
the reliance on who you need to be in the future will be relatively natural.
And as we all navigate this future, Shade, I think we're all very appreciative of the advice you've shared with us today and the story that you've gathered for us of Cloud Era's transformation and the impact that it has had already on the business.
I'm excited to hear more from your Insight Assassins.
in the coming months and years and to see how engineering roles continue to evolve within Cloudera and with the data and analytics teams.
Do you have any closing thoughts for our audience that you'd like to share?
Yes, I think there's a way that we can perform, conduct and leverage AI safely.
Today we talked about it in terms of a business and setting up environments and sandboxes and air-gapped capabilities to do it.
But I do want to encourage your audience, given recent news, that humans are still in charge.
Governments are the new way in which the French Revolution takes place now.
It's violence-free, but it is governments who will act on our behalf to keep AI safe across the world and safe within this country.
And so I try not to overreact to the news items this week and last week.
And I still have my faith in my fellow humans and I can see a pathway where this is all going to be good and we're all going to thrive in the age of AI.
And one final question, if you don't mind.
As we think about this transition, folks who have senior level skills, who have deep knowledge are often excelling.
They're able to be positioned for those next steps.
They have leverage points, but many folks who are just entering the workforce or are more junior are finding themselves sometimes struggling or simply experiencing a lot more fear.
What advice would you give to those folks who are getting started and maybe taking on their first month at a new job or looking to enter an industry?
I actually provide this advice on probably a biweekly basis.
There are many people who have enabled my career and now their children are fresh out of school and unemployed and I am returning favors by educating those folks and mock interviewing them and guiding them through what they should do.
And I think the advice that I've given that's resonating the most is cloud code would be great, but if you can just do cursor, if you can just do chat GPT, whatever your capability is technically, automate your last job.
What are the things that you did?
Generate synthetic fake data.
Guide AI through the actions of what that job was.
Take on in this time in your life these projects wherein you can understand and work with AI.
Because when you finally get to my desk and I'm interviewing you for a position and you say to me, oh, I did these 27.
projects myself.
Oh, tell me about those.
Oh, I automated my last job.
I did an app that does this.
I had AI do this.
That's going to be extremely compelling to me that you came, even though it's not on the job experience and you understand how AI is changing the world and you are using it to make your world better.
That's the number one piece of advice.
I've talked to a couple of early stage developers who are in college right now and are worried about what they should be doing.
And I've been telling them, look, build in public, you know, go contribute to an open source project, you know, rebuild something interesting.
Like people want builders.
They want people who are eager to learn.
And there is more opportunity to do that today than ever before.
And while there are challenges you're going to have to navigate, there is so much opportunity to just go out and make an impact if you have the chutzpah and want to spend the time on it.
So Shade, really appreciate you taking the time to walk us all through your approach and perspective and listeners make sure that you drop a comment on YouTube, on Spotify, wherever you're listening.
We'd love to hear from you.
We'd love to know what else we should have asked Shade.
We can probably get him back or drop him a couple more of our email if we need to.
So Shade, thank you so much.
Really appreciate you taking that time.
Thank you, Connor.
Awesome.