You are Early to the Data AI Party

Don't worry, everyone is about to show up

Being an early adopter of technology is a skill. It means knowing when to look around the corner and see what is coming next. That mindset helped me build a career in data analytics, both as a technical individual contributor and as a product manager for data products. I like testing new tools and looking for moments that shift how we work.

Even with that mindset, I have found it hard to keep up with the pace of AI.

We used to call this all Machine Learning. In the late 2010s, most people in the field avoided the term “AI” because it sounded inflated. What we were building were systems that generated statistical outcomes from data. We ran image recognition demos, experimented with chatbots, and shared projects on LinkedIn. You could learn something new, apply it at work, and expect it to stay relevant for at least a few years. It was hardly “Artificial Intelligence.”

Then in 2022, ChatGPT changed everything.

By 2023, people were sharing diagrams of transformer architectures. Retrieval-augmented generation (RAG) systems were showing up in blog posts. Developers started using agents to complete tasks. AI began writing full blocks of code with a single keypress.

In 2024, something called Model Context Protocol (MCP) appeared. It added complexity and created confusion about how to structure LLM applications.

Now in 2025, people are asking if vector databases are still needed. There are questions about whether tools like Cursor will survive after their funding runs out. Developers are comparing command-line agents. Semantic layers are being talked about like they were just invented.

All of this has happened in just over two years. And it still feels like the beginning of a party that has only just started.

If you have delivered a RAG system, implemented an MCP server, or built your own AI agent, you are already at the party. If you have used those tools to solve a real problem, you are at the party too.

You are here early, but more people will be showing up soon.

Programming was the first and most obvious target for LLMs. These models are trained on large amounts of unstructured text and code, so it made sense that software development was the first use case to take off. More engineers are using LLMs to review, write, and refactor code. Many are still skeptical, but they will show up to the party eventually.

Data has been harder to integrate into these systems. That is because data usually has structure. It is built for analysis, and analysis depends on consistency. That makes it more difficult for probabilistic systems like LLMs to produce reliable outputs on their own.

But it turns out we already had the tools we needed. Instead of using LLMs to write raw SQL from a prompt, we are now using semantic layers that define metrics and dimensions. These layers give the model context. MCP servers guide the model toward the right query by enforcing structure. Together, they reduce the freedom of the LLM but improve the reliability of the result.

This is the party that data professionals want to attend.

Talking to your data is not just hype. It is not about replacing SQL. It is about guiding people to the right answer through systems that understand what the data actually represents. It means creating strong definitions, clear contracts, and shared understanding across teams.

More Software Engineers are starting to care about the data lifecycle. More Product Managers are asking questions without triggering a long thread in a private analytics Slack channel. And when they finally show up to the party, the answer is already waiting.

Your CEO will eventually be there. Your marketing team will eventually be there. Your manager will eventually be there.

And the best part is that you are the one who threw the party. You saw what was coming. You tested the tools early. You set the foundation before most people even knew what this was going to be.

The Data AI party is just beginning. And it is going to be big.

---

If you are looking to start your own Data AI party you can check out a couple of my links here:

Connect Cursor to BigQuery MCP
Check out Rill’s MCP server in action