A pattern I've seen is that notebook are great for one-off EDA and report automations, but if you start automating a lot of work (ie: three end of month reports) then you start wanting to have shared utilities, a dashboard to share with colleagues, etc. Jupyter isn't good at multi-file projects. So you end up moving to vscode and into .py files
The iron man helmet, but for spreadsheets.
A proactive AI that:
> calls your attention to incoming missiles (data quality issues that will blow up your analysis)
> gives you readily accessible super (builds the most useful data visualizations, separates missing data into a separate file for manual review, etc.)
There's so many details you need to pay attention to when building a real-world, real-stakes report automation. Our goal is to offload the mundane stuff to Mito AI so you can focus on the highest leverage work that you are uniquely skilled at: making sure the analysis handles your specific business context with 100% accuracy.
@helloiamleonie OpenAI, Anthropic, Google SDKs for making inference requests routed through AWS lambda functions, and our own harness to orchestrate it all for data science workflows.
Took a few refactors to get the harness right but now it works wonderfully and easily extendable
The Citibike doc at the Navy Yard wework usually has an attendant so that it can hold ~130 bikes instead of it normal 30 bikes. Today was the dreaded 1 day of the quarter when she was not there, sending 100 people to race to the nearest docs.
Would've loved a text from Claude Code this morning to leave the house extra early to make sure I got a nearby doc.
@0xyorke@mattmurrs I want my agent to send a drone ahead to the CitiBike doc w/ one remaining ebike to scan the bike out and keep it safe until I get there
@leerob any advice here? It's getting incredibly hard to review agent changes. Notice it says 3 lines changes in the Agent taskpane, but the file is lit up with diffs all over the place
Cursor's review has been showing me inconsistent diffs. Notice here:
1. The Agent Taskpane review says two files were changed with only a few lines each
2. In the editor it's showing me way more new lines than the Agent reports (These are old, already committed changes that I made yesterday with a different Cursor agent)
3. In the editor toolbar it shows me that 13 files were edited.
This all started, when I edited an svg file and now if I ever click the "Review Next File" button after I click "Keep File", it takes me to that svg file.
Is there any way to reset my change history / fix this issue? It's making it very hard to review the agent's work! @leerob@cursor_ai Thanks for your help!
Here is my build:
Version: 3.0.16
VSCode Version: 1.105.1
Commit: 475871d112608994deb2e3065dfb7c6b0baa0c50
Date: 2026-04-09T05:33:51.767Z
Layout: editor
Build Type: Stable
Release Track: Default
Electron: 39.8.1
Chromium: 142.0.7444.265
Node.js: 22.22.1
V8: 14.2.231.22-electron.0
OS: Darwin arm64 25.3.0
Being able to easily share these flows with the rest of your team is a huge deal.
When we onboard new teams to Mito, usually one person takes on the role of context creator. They:
1. write up custom rules like: If you want churn data, merge these tables on these columns, etc.
2. try to teach the rest of their team how to use the custom rules, how to keep them in sync, etc.
For teams that already use Notion, some of that context, and all of that collaboration piece is already taken care. Huge ops win.
.@abacusai, @arvinds can you confirm how to set the response format for the LLM router:
- /app/route-llm-apis says to use {"type": "json"}
- /help/developer-platform/route-llm/api says to use {"type": "json_object"}
@arvinds@abacusai Appreciate it. Do you know if that response type is passed through to non-openai models. I opened an issue here https://t.co/zFOUWpNtSq
Anyone else noticing that gpt-5.2-codex is prone to returning the complete response_format object twice in one completion?
Haven't seen this issue with other models from openai
Getting Cursor to verify its own changes is such an unlocked workflow, but recently I've been having some issues getting Cursor to use the browser.
Any ideas on what I'm doing wrong here @leerob? Appreciate any help greatly!