AI is changing how our data team works, and that’s great

When we look back, we'll see 2025 as the year data roles really started changing. As the AI capabilities in Hex and other products have started to mature, the work we're doing has started to evolve, too.

I’ve been leading Data at Hex for the last few months, and have spent a lot of my time trying to make sense of this, and where it's going – and come to believe that this is great for us, and data teams everywhere.

But it can be hard to see sometimes. With change comes uncertainty, and with uncertainty comes anxiety – even a touch of existential dread – and many data practitioners and leaders are feeling that. But I don’t see AI analytics as a threat to data teams; this is a technology that’s genuinely helping us get closer to the goal of helping our organizations make better, faster decisions with data. It will, however, require us to evolve how we work and to think differently about where we add value.

The Hex team has been dogfooding our Notebook Agent and other early features for a few months, which let us live in the future a little bit. I’m not prone to hyperbole about tools, but we’ve quickly seen how this kind of AI-assisted workflow for analytics has changed how our team works and what we prioritize.

Here's what we're learning so far about navigating the role of the data team in the age of AI.

1. Self-serve is actually happening

The team at Hex is more data-savvy than most organizations I’ve worked in, but even then, only a small portion of our company actually created their own novel analyses six months ago. The rest came to the data team to partner with us on building apps and generating insights.

But now, with our new AI tools, many more people are able to do things themselves. As it turns out, natural language really does lower the barrier.

They’re not just answering simple questions – my colleagues are getting into way more depth than was possible before. It's been interesting to watch this from my seat — it's sometimes incredible, sometimes frustrating, and unlike anything I've seen in my career.

We’ve got our Head of Revenue investigating what actions organizations take in the product before they churn. Our Support Team leader is digging into customer ticket data to understand the temporal distribution of ticket creation to make decisions on staffing. Our Head of Product is analyzing how customers move through pricing tiers, what features they use, and how their teams grow over time to shape the evolution of our pricing strategy.

These analyses would have required an analyst only months ago, but now we only even know they’re happening when folks tell us or ask us to check their work.

Our data team is still doing analysis, but it’s more targeted, strategic, and more focused on creating reusable frameworks or jumping-off points for our colleagues to go even further with the agent (and we think this is great!).

2. We need to invest more in infrastructure and less in data-team-generated analysis

In the past, infrastructure investments were harder to justify because a clever analyst could bridge foundational gaps. Suddenly, a solid foundation with fewer “gotchas” is much more important because it serves so many people via the agent. And at the same time, a lot of QQs no longer cross the threshold of the #data channel as folks self-serve with the agent.

In many ways, this is a dream for me. Instead of spending a big chunk of our team’s time on one-off requests, we are focusing on the infrastructure that makes all analysis better.

Our semantic models are becoming critical for consistency and quality. We've invested a ton of time in testing various semantic tools, including early Hex-native features, as well as experimenting with the structure and coverage of our SL. Documentation and rules files provide additional context that AI agents need to generate reliable insights. While we may have let docs slide in past years, we are now very careful to ensure we have good coverage and up-to-date definitions of our data.

Analysis itself is becoming more ephemeral. When it's easy to generate insights on demand, the lasting value shifts to what's behind the curtain — the context, definitions, and infrastructure that make those insights trustworthy.

3. Analytics engineering has always been context engineering, and it’s more important than ever

Folks are talking about "context engineering," and designing systems to provide the right information to accomplish tasks.

But that’s not new to data teams. We have always been the translation layer between raw data and business decisions. Now, we're going from building those systems to enable a small number of humans to generate insights, to unleashing many more people via agentic analysis.

Think about what you actually do when someone asks why revenue dropped last month. You don't just write a query. You know Q4 always looks weird because of seasonality. You remember that 2019 acquisition that still throws off year-over-year comparisons. You understand which data sources to trust and which ones are held together with duct tape. You translate business questions into the right analytical approach and turn numbers into narratives that actually mean something.

Now we have to scale that context to serve not just a handful of stakeholders, but an entire organization working alongside AI. This is, fundamentally, just analytics engineering. The core skill of analysts and analytics engineers has always been connecting business strategy to technical details. The need for that human skill set isn't disappearing — it's becoming the foundation for everything else. There’s enough meat here for an entire separate post - coming soon.

The work will keep evolving, and we get to shape how

Whether you are an AI skeptic, aficionado, or anything in between, your stakeholders and organizations are going to pull you into this future. The change can feel intimidating. When you hear "AI can do analysis now," many data practitioners worry about becoming obsolete. But this shift creates opportunities to work at a higher level and focus on the parts of data work that are genuinely fulfilling and uniquely human.

The choice isn't whether to embrace AI analytics. The choice is whether you'll build the foundation that makes AI work well for your organization, or whether you'll let people navigate that wilderness on their own. Building robust context layers, investing in solid infrastructure, and focusing on the human skills of translation and judgment turns your data team into a force multiplier. Your work can now serve more people, drive more decisions, and create significantly more value than ever before.

The data team has always been about turning information into intelligence. Now we get to do it at scale, with the chance to make our organizations genuinely data-informed. If we approach this transition thoughtfully, we can build something better than what came before.