Thoughts on Data Agents

This week, I spent quite some time exploring what a data agent could realistically offer—especially when it comes to replacing or augmenting parts of an analyst’s workflow. Here are a few things that stood out:

1. Domain knowledge isn’t optional—it’s everything.

No matter how fancy the model is, if the agent lacks domain context, the output becomes generic at best, misleading at worst. Giving agents access to product logic, metric definitions, and business rules isn’t a nice-to-have—it’s a prerequisite.

2. Agent chaining feels like the future.

Instead of asking a single agent to do everything, we can chain smaller, specialized agents together.
Think: one agent retrieves data, another analyzes it, and a third generates a chart or report.
This modular approach brings us closer to truly automated, end-to-end workflows.

3. Long-term memory is underrated.

If we want agents to generate value over time—not just in isolated tasks—they need persistent memory.
That means remembering previous analyses, knowing the user’s preferences, and building up context over weeks or months.
Without that, every session starts from scratch.

4. Notebooks + Agents = Powerful automation.

Imagine scheduled agents that run notebooks, convert results into HTML, and send them out automatically.
You get consistency, traceability, and no more forgotten reports.
This could be huge for any team with regular reporting needs.

5. Different roles, different expectations.

One thing became clear: people want different things from an agent depending on their role.

• PMs want links they can trust—not SQL queries.
• Data Scientists want SQL generation, smart slicing, and maybe a shortcut to writing the same report again.
• Engineers want a quick way to verify if tracking is working, ideally via auto-generated SQL.
• Leaders want a clean, high-level summary. No jargon. No noise.

6. Integration is important.

Even the smartest agent is useless if it’s not accessible.
Embedding it into existing tools—dashboards, conmunication apps, whatever—is how you make it part of the workflow.
Without that, it’s just another tab people forget to open.

I’m still figuring things out, but this week's deep dive made one thing clear:
Agents won't replace analysts anytime soon—but they might just become our most valuable teammates.