Skip to main content

2026-08 v1.7

Your notebook just got a lot more capable.

Until now, each cell worked the same way: you wrote a prompt, the AI responded, and you moved to the next cell. It was fast, but it was on you to string the steps together — plan the analysis, decide what to run next, and check whether the output actually made sense.

That changes today.

Agentic Notebook

What's new​

The notebook now works agentically, end to end. Instead of executing one prompt at a time, the AI can:

  • Plan — break down your goal into a sequence of steps across cells
  • Orchestrate — decide what to run, in what order, and adapt the plan as new information comes in
  • Execute — run code, transform data, and produce outputs across the notebook
  • Evaluate — check its own results, catch errors or dead ends, and self-correct before moving forward

In short: you describe the outcome you want, and the notebook figures out and carries out the path to get there.

Why it matters​

Multi-step analysis used to mean manually chaining cells together, re-running things when an earlier step turned out wrong, and eyeballing outputs for mistakes. Now the notebook handles that loop itself, planning ahead, adjusting mid-flight, and checking its own work, so you can focus on the questions you actually care about instead of the mechanics of getting there.

What stays the same​

Every cell is still visible and editable. You can inspect what the AI planned, step in at any point, tweak a cell, or take over manually — the agentic workflow doesn't hide anything from you, it just does more of the legwork by default.

Get started​

Open any notebook and give it a goal instead of a single instruction — for example, "clean this dataset, find the top drivers of churn, and chart them" — and watch it plan and work through the notebook on its own.