We've made a bunch of changes in Posit Assistant with approval fatigue in mind: * Many obviously-safe commands will be automatically approved by the system. * /auto will have another model approve/reject any commands that would otherwise require your attention. assistant.posit.co/docs/feature...

Just wrapped up our first session, "AI, LLMs, and Big Data in R" with Kristin Bott, Ed Borasky, and Kylie Bemis"! We learned about using LLMs for coding in R,ethods for code parallelization, Posit and R tools for memory efficiency, and more. #LLMs #CascadiaRConf #AI #Posit #rstats #R #coding

Interested in learning how to contribute to open source? We'd love to have you. :)

It seems like LLMs are getting much better at faithfully describing data visualizations that show surprising trends. With @sara-altman.bsky.social on the @posit.co open source blog: opensource.posit.co/blog/2026-06...

Congratulations to the R Core Team! 🎉🎉🎉

How can you get better R code out of Claude, ChatGPT/Codex, Gemini, and other generative AI? These tips have worked for me: www.infoworld.com/article/4184... #Rstats #R #GenAI #LLMs

I want to publicly thank @jcheng5.bsky.social for being willing to write this post, laying out how our AI tooling has evolved (including those eyebrow-raising product names)! 😅 opensource.posit.co/blog/2026-06...

@sara-altman.bsky.social and I will be speaking at the R/Pharma GenAI Day tomorrow about agents for data science! The talk is free, virtual, and should also feel relevant for data folks outside of Pharma. :) Register: events.zoom.us/ev/AkKYQ7Ury...

Coding agents like Claude Code are great at general-purpose work. Posit Assistant is built specifically for data: #RStats / #Python session access, plots in the chat, and specialized data analysis capabilities. Post + video: opensource.posit.co/blog/2026-06...

Fable consistently devolving into incomprehensible shorthand over the course of a conversation is the weirdest experience

Fable 5 is the new highest scorer on bluffbench! The eval measures models' ability to accurately describe plots that show counterintuitive patterns. Six months ago, the strongest models were still in the single digits. simonpcouch.github.io/bluffbench/

The user interface in the new Posit Assistant is one of the best and most intuitive I've seen for using AI tools for editing code. This will be excellent for students and researchers already familiar with RStudio. This is a brilliant harness and user experience for using AI for data science 🎉

We admit we're in an awkward spot right now 😅 in this migration for our AI coding assistance tooling, but I am really happy with how the Posit Assistant experience is shaping up. Give it a try, if you haven't yet!

I've just tried Posit Assistant on the free plan and must say, I'm impressed. Using the super-cheap Gemma 4 26B model (10% cost of Sonnet) on free plan it worked instantly. Hats of to the #posit team for integrating this into RStudio and allowing people to test with free plan!

Kickstarting June the only way we know how: with incredible open-source updates. We're excited to share recent #RStats and #Python updates to make your life easier, from mirai, mori, Shiny, Quarto, to events and resources, and more. Read the Glimpse newsletter: opensource.posit.co/blog/2026-06...

Announcing mori 0.1.0: Zero-copy shared memory for #RStats! Parallel R usually means duplicating data across every worker, causing RAM bloat. mori fixes this by sharing R objects across processes via OS-level shared memory. Share once, read anywhere Learn more: opensource.posit.co/blog/2026-04...

Re-ran this eval against Opus 4.8, Gemini 3.5 Flash, and GPT 5.5. Opus 4.8 is a modest improvement over the previously tested Opus models, but Gemini 3.5 Flash is the real stand-out! simonpcouch.github.io/bluffbench/

Sarcastic gradient descent

Given the recent burst of activity around enterprise pricing and contracts, I think April 2026 was the month when both OpenAI and Anthropic found product-market fit simonwillison.net/2026/May/27/...

Posit Assistant now ships with a self-knowledge skill! The skill teaches the agent about its own interface and configuration options. Here's an example of the agent configuring itself with an MCP server:

Using LLM chat in a browser interface is just the beginning. You can unlock so much more power programmatically! At Pyladies Monterrey, I spoke about chatlas, a lightweight library that connects #Python with LLMs. Check out the slides and summary (in 🇪🇸 and 🇬🇧): ivelasq.rbind.io/talk/connect...

¡Qué keynote de lujo! 🤩💫 Si te faltaban excusas para venir a #LatinR2026, mira quién será una de las personas que darán las charlas principales. Además, todavía estás a tiempo de mandar tu charla/póster: openreview.net/group?id=LAT... #RStast

In this edition of the AI Newsletter, we break down rising AI token costs from frontier model providers and announce the release of Gemma 4 to Posit AI as a cost-conscious alternative. 🔗 Read it here: opensource.posit.co/blog/2026-05...

In the last year, @posit.co has released one agent called Positron Assistant and another called Posit Assistant, which is kinda confusing. @sara-altman.bsky.social and I sat down to talk about how these agents compare and where they're headed: www.youtube.com/watch?v=Y9P2...

I have to update one of the most inflammatory things I said back on the bird site, but for the LLM era: 🔥 🔥 🔥 🔥 If your agent prefaces every tool call with `cd /the/current/workspace &&`, I will come set your data center on fire 🔥 🔥 🔥 🔥

AI is expensive, but we're doing our best to make it available as cheaply as possible.

Gemma 4 is now available for Posit AI subscribers! We've priced the model at a tenth of the per-token cost of Claude Sonnet, and are really impressed with its performance for quick data analysis tasks. posit.co/blog/gemma-4...

We’re excited to launch the Posit Open Source website! 🚀 We’ve unified 15+ years of tools, 900+ blogs, and 1,600+ videos in one place. Explore it here: opensource.posit.co Huge thanks to the team: @cwick.co.nz Greg Swinehart @hadley.nz @ivelasq3.bsky.social @jeroenjanssens.com

vitals 0.3.0, an #rstats package for LLM evaluation in R, is now on CRAN! Among other things, this releases introduces a new `generate_structured()` helper, allowing users to evaluate LLM systems' effectiveness at extracting structured data. vitals.tidyverse.org/news/index.h...

Looking to connect with Posit folks on Bluesky? 🌐 We've updated our Posit Starter Pack! If you haven't looked at it lately (or are seeing it for the first time), discover and add some incredible data scientists, developers, and team members to your feed. bsky.app/starter-pack...

I'm bringing my substack back to life to talk about AI and data science: open.substack.com/pub/tidydesi.... I have conflicted feelings about both AI and writing about AI but I want to try and work through them in the open, and I hope you'll join me on this journey.

90% of data science is...

This release has some nice ergonomic improvements for #rstats users querying Snowflake or Databricks. I'm excited about the potential of the workload identity functionality in particular.

Posit AI is $20/mo, most of which goes to model API costs. Does that mean other, subsidized subscriptions provide more tokens at the same price point? @sara-altman.bsky.social and I wrote about the tokenomics of Posit AI and the end of token subsidies: posit.co/blog/posit-a...

odbc 1.7.0, an #rstats pkg that allows users to connect to ODBC databases, is now on CRAN! This release tightens integration with Databricks and Snowflake, in addition to a few provider-agnostic bug fixes. github.com/r-dbi/odbc/r...

This is excellent. I particularly like the definition of the "Zombie Internet", which starts: "It’s people talking to bots, people talking to people, people creating “AI agents” and then instructing them to interact with people. It’s people using AI talking to people who are not using AI [...]"

It's not [A], it's [B]. Here's why that matters. No [C]. No [D]. Just [E]. Whether you [F] or [G], [take some action].

The newest release of Posit Assistant, an agent for coding and data analysis, includes a "data cleaning mode." When enabled, the agent will run quality checks and surface decisions about e.g. import issues, factor levels, etc to the user. In the AI Newsletter: opensource.posit.co/blog/2026-05...

So do I need Google Gemini Enterprise Agent Platform or Google Gemini API or Google AI Studio or Google Vertex AI or Generative AI on Google Cloud Good gosh just give me an API key and let me pay you for tokens

I’m presenting some data findings about funding for freelance journalists at the IIJ Town Hall. Come through! 👀 📊

roxygen2 8.0.0 out now: opensource.posit.co/blog/2026-05... ! So many goodies in this release including S7 support, tons of R6 improvement, a new config style, fewer dependencies, and so so many bug fixes #rstats

Still pretty wild to me that this works (and quite well!)—pull a pre-trained neural network, provide some of your own data in-context (in lieu of "training"), and start making predictions. Kudos to the team for the R interface and tidymodels integration.

Out of curiosity, asked Opus 4.7 for feedback on a blog post draft. Some of the feedback on broader structure was reasonable, but every concrete edit suggestion was just introducing an LLM-ism—"not X, it's Y", excessive line breaks, overformatting. 🥴

Seems like the party's over on token subsidies. Anthropic and OpenAI clamping down on subscription rate limits, Copilot transitioning to usage-based billing. Knew this would come some day, but did not suspect in the same month that laptop-size models could start convincingly driving coding agents.

Four engineers behind Posit Assistant broke down how to build an AI agent for data science — reasoning, agentic workflows, modeling guardrails, and chart extraction. Watch: www.youtube.com/watch?v=A2sa... Download RStudio: posit.co/download/rst... #PositAI #RStats

Our AI newsletter has moved to the new @posit.co Open Source blog! In this edition, AI and cybersecurity, a step change for local coding agents, and a new dataviz engine for SQL. opensource.posit.co/blog/2026-04...

I've been working on this over the last few months and in my opinion it is just beyond cool. Grammar of graphics for SQL users!

If I might translate this: you can run an LLM on a cheap (~£300, e.g a Raspberry Pi5 with AI Hat+ 2) computer you own, in your house, disconnected from the internet, which will never give your data to Sam Altman or Elon Musk or Mark Zuckerberg, and will outperform ChatGPT from a year ago

I'm planning on a GIANT roxygen2 release in the near future (tons of bug fixes, improved R6 support, new S7 support, ...) so if you use for your packages, I'd really appreciate you trying it out and letting me know if you see any problems! github.com/r-lib/roxyge... #rstats

A few months ago, any LLM that I could run on my Macbook scored 0% on an agentic coding eval I put together. This month's Qwen 3.5 and Gemma 4 releases both scored 90%. On my blog: simonpcouch.com/blog/2026-04...