Yeah vibe coding your own platform required a lot of architecting, thinking, and tokens. Basing off of something like streamlit may be much more realistic.
If you're in gcp, data studio can be a great option. I'm using it at my current company and I'll continue using it as long as it makes sense. Even when we introduce another tool, wed keep data studio in our tool disposable and use it when it makes sense
AI is not replacing BI, but it’s dramatically devaluing tableau. Locking metrics at the BI level should have never been the best practice. Same as the lack of documentation on data definitions. Fortunately AI is pushing for both and making both easier to implement.
Definitely. You should back these into a database ASAP. And if you're vibe coding with measures defined in apps, for the love of God at least add them to a markdown file for reuse at a bare minimum.
Great read. You’re spot on about the pricing disconnect. Charging high per-seat costs just to build basic dashboards is a dead model. But as you noted, the answer isn’t abandoning BI entirely; it’s merging traditional BI governance with AI agility to unlock actual self-service directly on the warehouse.
The 'vibe coding' pilot is the perfect way to test the dilemma. It gives a reality check to leadership of what is possible with AI. For simple data, a custom AI layer might work. But the second an org needs RBAC, caching, and real security, rolling your own becomes an operational nightmare.
This is exactly what we’re seeing on the ground. Enterprises want to slash costs but won't compromise on trust and the market mandate is clear: the Tableau exodus is real, but buyers demand the new stack actually earns its keep.
> The challenge is all the other stuff BI does - security, auditability, content management, reusability, load balancing, caching, distribution, metrics stores - it’s a big list. AI can already make a good dashboard, it can’t automatically personalize a report and send it to 1,000 people every Monday morning.”
This is always the part that gets skipped. In demos and PoCs, we always just test the dashboards and the cool stuff. In real workflows, these parts pop up as issues far more often but execs who aren’t on the hook for this part basically can’t see it and so doesn’t account for it.
On cost, I had to do an Omni evaluation to replace Tableau myself. At our size, we are not going to save ANY money by switching to Omni. But now it’s nearly halfway through the year until our license expires. Wish us luck!
Good analysis, Ryan. The Tableau exodus isn't just a sudden trend; it’s an acceleration of traditional BI’s core failure. For decades, companies have thrown millions at these tools plagued by horribly low user adoption rates. What is exciting is that AI is finally treating semantics and metrics as first-class citizens (although I don't think the vibecoding crowd has picked up on that yet). I expect the pendulum to swing back towards governance, security, and trust ... hopefully.
Very nice piece! I learned a lot.
Thank you Justin!
Yeah vibe coding your own platform required a lot of architecting, thinking, and tokens. Basing off of something like streamlit may be much more realistic.
If you're in gcp, data studio can be a great option. I'm using it at my current company and I'll continue using it as long as it makes sense. Even when we introduce another tool, wed keep data studio in our tool disposable and use it when it makes sense
AI is not replacing BI, but it’s dramatically devaluing tableau. Locking metrics at the BI level should have never been the best practice. Same as the lack of documentation on data definitions. Fortunately AI is pushing for both and making both easier to implement.
Agreed. Composability is the solution.
If you thought measures in a BI tool are hard to figure out, wait until you vibe code it.
Definitely. You should back these into a database ASAP. And if you're vibe coding with measures defined in apps, for the love of God at least add them to a markdown file for reuse at a bare minimum.
Tableau is the new Oracle of analytics. Feels like they have reached their zenith and it's all downhill from here.
Tale as old as time.
Great read. You’re spot on about the pricing disconnect. Charging high per-seat costs just to build basic dashboards is a dead model. But as you noted, the answer isn’t abandoning BI entirely; it’s merging traditional BI governance with AI agility to unlock actual self-service directly on the warehouse.
The 'vibe coding' pilot is the perfect way to test the dilemma. It gives a reality check to leadership of what is possible with AI. For simple data, a custom AI layer might work. But the second an org needs RBAC, caching, and real security, rolling your own becomes an operational nightmare.
This is exactly what we’re seeing on the ground. Enterprises want to slash costs but won't compromise on trust and the market mandate is clear: the Tableau exodus is real, but buyers demand the new stack actually earns its keep.
> The challenge is all the other stuff BI does - security, auditability, content management, reusability, load balancing, caching, distribution, metrics stores - it’s a big list. AI can already make a good dashboard, it can’t automatically personalize a report and send it to 1,000 people every Monday morning.”
This is always the part that gets skipped. In demos and PoCs, we always just test the dashboards and the cool stuff. In real workflows, these parts pop up as issues far more often but execs who aren’t on the hook for this part basically can’t see it and so doesn’t account for it.
On cost, I had to do an Omni evaluation to replace Tableau myself. At our size, we are not going to save ANY money by switching to Omni. But now it’s nearly halfway through the year until our license expires. Wish us luck!
I always tell people that demos have never been more fake, it's so easy to vibe code whatever the prospect asked for in discovery.
I feel for you my friend - good luck! Let us know how it turns out!
Good analysis, Ryan. The Tableau exodus isn't just a sudden trend; it’s an acceleration of traditional BI’s core failure. For decades, companies have thrown millions at these tools plagued by horribly low user adoption rates. What is exciting is that AI is finally treating semantics and metrics as first-class citizens (although I don't think the vibecoding crowd has picked up on that yet). I expect the pendulum to swing back towards governance, security, and trust ... hopefully.