Vibe Analytics: the new era of data analysis?
Since February 2025, vibe coding is everywhere. Of course, we focus on data and analytics here, so a natural question comes to mind: is Vibe Analytics the next thing in data?
Please, try not to vomit. At this point, you’re probably fed up with hearing the term Vibe Coding and the mere mention of Vibe Analytics makes you want to scream.
For those who haven’t heard, Vibe Coding was coined in February by Andrej Karpathy. It’s basically the idea of having AI do coding for you, based on input from you, and hopefully you get the result you want. Also, hopefully, the result is a solid piece of software, that works well, doesn’t have security holes, and prints money.
While it’s great for prototyping, or building some apps to help with your Mexican breakfast, the current state of affairs is that you shouldn’t rely on it for actual production software. Just like you shouldn’t rely on AI to write your entire court brief.
So what does Vibe Analytics mean? I would define it as:
Using AI to explore data for analyses purposes.
Does it mean that you can just ChatGPT-away your entire analyses and blindly rely on AI for data analysis? Not yet. Note the use of the word “explore” in my definition.
But it’s also not complete nonsense either.
Where is this going?
AI has proven itself quite useful in analytics tasks. For example, you can upload a CSV into any of the chatbots (just make sure your security team is OK with it 😅), and get AI’s help to explore the data.
You can even use AI to write SQL, Python and R for you.
But, can you just straight up ask a natural language question and AI will understand exactly what you need, find the right data, visualize it the right way and serve it on a silver platter? Not yet. Databricks, Google and Snowflake sure are trying, but our conversations with over 200 organizations show this is still quite elusive.
Why?
AI is terrible at understanding your full business context without a lot of work by you.
AI hallucinates (as we all know) and grounding it in a semantic model, data profile, and past SQL queries is necessary to reduce this. (even then, not 100%)
Building a working search mechanism is much harder than it seems.
Different personas need different interactions with AI. For example, consider a business user with no analytics background, vs a 5-year analyst, in the same business department.
So while Vibe Analytics is a cute term that can cause eye-rolls, it is pointing to a very clear and obvious goal the entire industry is aiming towards:
Making data and analytics much easier for humans to produce and consume.
The number 1 issue - trust
Just like Vibe Coding generates untrustworthy applications, Vibe Analytics can potentially generate “insights” that are dangerous to rely on. Sometimes, you just need some basically directional data, in which case this level of analysis is ok. But, often times, you’re relying on the data and analysis to make an important business decision.
And when making that decision, you need to be able to trust the insight, the process that generated it, and the data it relies on.
Companies, including Solid, TextQL and Jedify, are building end-to-end solutions to achieve this by leveraging the existing data stack. Others, like Hex, are building an analyst notebook with AI capabilities that directly connects to your data warehouse. Then there are data observability solutions like Monte Carlo.
Hundreds of teams are chasing the goal of providing trust-worthy AI-driven analytics right now. It’s truly an exciting time to be in this space. The winners will be those who can build a reliable platform, where AI is very useful, the trust issues are either removed or made obvious to the user and is easy to roll out within an existing data stack.
As my 13-year-old daughter loves to say - “we’re all just vibing”.