Blog

Data teams are racing to give agents the context they need. But the teams making the most progress aren’t buying another layer. They’re writing down what they know and keeping context close to the data, code, and work itself.

Most of the data work ahead will be done by agents. But where will that work happen?

Today, OpenAI launched the Codex data analytics plugin with a native connection to Deepnote. The business context, stored in your Deepnote workspace - the notebooks, scheduled analyses, data apps, and integrations your team already maintains in Deepnote is available to Codex directly, so any data explorations in Codex can start from your real work.

Reinforcement learning becomes central once a model is expected to act, not just generate. Real tasks are sequential decision processes with tool calls, partial observability, and long horizons. RL-based alignment work frames this as an optimization loop dependent on interaction data, reward signals, and reliable evaluation.

AI agents are failing because they lack business context. The industry is scrambling to build "context layers" as a fix. But if your team works in notebooks, the context layer already exists - inside the analytical work itself.

This piece will help readers decide if Jupyter AI is the right assistant for their notebook workflow. It will start with what the project actually is, when it was introduced, and who is driving it. Then it will show how the chat UI and %%ai magic command work in practice, the experience, and where it stalls. The close will compare that flow with Deepnote and a real notebook example to judge which tool gets to a usable result faster.

We engineers hate writing tests. Fixing broken ones? Even less so. But for any enterprise-grade product, this is a must. So the true enjoyment of using AI in coding comes from using natural language and thinking models to address this issue. It all started with a regression after a release.

AI data visualization tools aren’t about who draws the nicest chart. They’re about how they handle definitions, calculations, transparency, and sharing when it counts. In this article, we run the same small benchmark across top tools to see how they compute churn, explain changes, show their work, and package results. The goal isn’t to crown a winner. It’s to see how hard each tool makes it to ship a wrong answer.

Event‑study walkthrough showing how adverse drug event signals map to pharma stock moves, combining price data with FAERS counts to measure market reactions. A practical framework for analysis, with a clear focus on method and interpretation.