I want to write about Codatum, a BI tool combined with an AI agent that I've been relying on heavily at work lately.

In data-driven work, the friction of "not being able to test a hypothesis the moment you want to" is genuinely stressful.

You open a query editor, figure out the SQL, pull the results, hook it up to a visualization tool, build a chart, then maybe paste it into a slide for discussion — and you repeat this cycle every single time. The overhead of producing the data becomes a barrier in itself, even though data analysis is really just a means to an end: the real work is the discussion and decisions that come after.

Since adopting Codatum, though, that friction has dropped considerably. In this post, I'll explain why we moved away from Tableau, what has actually changed, and what challenges still remain.

What Is Codatum?

Codatum is a next-generation BI tool built around the concept of "making advanced data exploration the norm — a data notebook for teams."

The company behind it, CODATUM Inc., was founded in October 2023 as a spinout from Plaid, the company behind KARTE (a customer experience platform). The product carries deep expertise in data infrastructure that Plaid developed through KARTE.

  • Notebook format: SQL query results render in real time alongside charts and narrative text, all managed in one place.
  • Multiple team members can edit simultaneously.
  • Data is always pulled directly from the data warehouse rather than copied, so results are always fresh and accurate.
  • Supported data sources include BigQuery, Snowflake, Amazon Redshift, and Databricks.

Codatum also ships with an AI agent — Codatum AI — that integrates seamlessly and acts as an analysis expert tuned to your team's context.

That's the rough overview.

Why We Adopted It

At 10X, Tableau had been our official BI tool. Over time, though, Tableau's steep learning curve led to a proliferation of unofficial dashboards built to avoid it — a growing mess with no clear ownership.

When we set out to fix this, we identified three core problems:

  1. Balancing data governance with autonomy. Having a central data team handle every query request doesn't scale. PMs and BizDev needed to be able to self-serve.
  2. Enabling self-service analysis. We wanted BizDev to be able to run routine investigations entirely on their own. Waiting on the data team for every request killed speed.
  3. Improving the quality of collaboration. We didn't just want to share results — we wanted to share the queries themselves and the thinking behind them. Being able to answer "how did you pull that number?" without extra effort was the goal.

Codatum fit these requirements as essentially "a Notion-style BI tool where you can write queries" combined with robust access control.

We rolled it out in phases from late 2025 through early 2026: Phase 1 (infrastructure), Phase 2 (pilot), Phase 3 (company-wide).

How Text-to-BI Changed the Way We Analyze

After adoption, the biggest shift in day-to-day experience has been Codatum AI.

Here's what our analysis workflow looked like before:

  1. Open the BigQuery console, think through the SQL, write the query
  2. Look at the results, revisit the direction, iterate
  3. Finalize the output table, export to Looker Studio or a spreadsheet
  4. Build a chart, take a PNG screenshot
  5. Paste the image into Google Slides or Notion

The core problem with this flow is that the business-ready output almost always ends up as a chart image.

Challenges in Turning Analysis into Business Value

  • Once a chart becomes a PNG, the live connection to the data is severed.
  • When the underlying data updates next week, someone has to manually swap the image.
  • Queries live on individual machines and sharing them takes extra effort.
  • Visualization quality depends entirely on whoever made it.

Codatum AI changes this at the root. You type something like "show me last week's order hit rate by category as a heatmap," and the AI generates the SQL, runs a dry-run to validate it, executes it, and produces the chart — all in one shot, with no friction.

Analyses I used to give up on because the query was too annoying to write are now easy to try. That's a bigger psychological shift than I expected — being able to act on a "I'm vaguely curious about this" impulse without a big setup cost has genuinely increased how often I look at data.

Where I Actually Use It

  • For evaluating AIO (AI Order — a product that uses AI to support optimal order recommendations), I drill into category-level heatmaps dynamically to identify which types of orders are seeing accuracy drops. That kind of analysis used to take the better part of an hour; now I can have an initial hypothesis in minutes.
  • Executive briefing reports have moved from Notion-with-PNG to Codatum. Not having to manually replace images every time the data refreshes is a real relief.

What I'd Like to See Improved

The more I use Codatum, the more I see its potential — but also the walls that still need to be climbed.

Admin API, IaaS Integration, and Audit Logs

Right now there's no integration with IaaS tools like Terraform, so you can't manage configurations declaratively. An API exists, but combining it with IaC to manage settings in a version-controlled, reproducible way isn't yet possible. That means manual operations at scale, and the operational burden grows as usage expands.

Audit logs are available on the Enterprise plan, but exports aren't supported, and notebook content details aren't recorded — there are real limitations. Back in the Tableau era, we built external monitoring scripts to detect unauthorized sharing. In Codatum, building equivalent external controls is difficult. Tracing who shared which notebook with whom after the fact is limited — and for 10X, where large-scale sharing with enterprise retail clients is a core need, this leaves meaningful governance concerns open.

Notebooks Tend to Get Messy

Another thing that's become apparent over time is the risk of notebook sprawl.

Each notebook has a 10MB size limit, and performance degrades as the number of elements and pages increases. Codatum's own docs recommend keeping notebooks compact and starting a new one when things get large.

The flip side is that without deliberate rules around organization, notebooks multiply and eventually go stale. This is exactly the "rogue dashboard" problem we had with Tableau. To avoid repeating that mistake, we need to design notebook governance policies early.

External Sharing and Security

Looking ahead, we'd like to give our retail partners access to their own data environment so they can run their own PDCA cycles independently. But today, Codatum's external sharing capabilities have security limitations that make us hesitant to extend access to partners.

At the notebook level, external users can be invited with editor or viewer roles. The issue is that anyone with editor permissions can invite additional external users — which creates the risk of unintended sharing. Combined with the audit log limitations mentioned above, if a mistake happens, you may not catch it until the damage is already done.

Before we can extend access to partners, the security foundation needs to be stronger.

Conclusion — BI Democratization Is Becoming Real

Overall, the product's potential is very high — honestly, it's the product I'm most into right now.

The integration with an AI agent has dramatically lowered the barrier to actually looking at data. I've experimented with using generative AI for data analysis since the early days, but in terms of product integration and polish, Codatum is the first tool where it genuinely feels complete.

At the same time, I'm looking forward to seeing the admin API, notebook governance tooling, and external sharing security mature alongside the product. When those pieces are in place, extending access beyond internal teams to partners becomes a realistic goal.

My plan is to stay close to Codatum's product evolution while continuing to strengthen our internal operational foundation.