Posit PBC

If posit::conf(2026) has been sitting in your maybe pile, this is the week it stops being a maybe. Fourteen days from now, posit::conf(2026) kicks off! We’ll hear from clinical data teams from Houston Methodist, Michigan Medicine, and the VA, Shiny engineering from Boehringer Ingelheim, policy analysis from the United Nations, and the people who build the tools at Posit. Sara Altman and Simon P. Couch on a real tension in AI-assisted analysis: good analysis means slowing down and questioning conclusions, and most of today's agents are wired to keep moving. Christine Y. Zhang on building visualization and communicating uncertainty. Emily Riederer on the context data scientists require to do meaningful work. Wes McKinney on skills to be effective and successful in an agent-powered future. If the travel doesn't work, the virtual pass streams every keynote and session on September 15 - 16, adds dedicated talk tracks on the 14th, and now includes live captioning in French, German, Italian, Portuguese, and Spanish. From $99, with free needs-based tickets available. Either way, we’d love to see you there: conf.posit.co 

To MCP or not to MCP? MCP is having a moment right now, and with an uptick in interest comes an interest in questions, too. -How can I set up an MCP server in a regulated environment? -When does MCP make sense, and when does it not? -How should I think about credentials and access? -Where should I host them? Our team recently held a webinar that broke down all these questions and how to host MCP servers using Posit Connect, and the recording is now up on YouTube! Link available in the comments 🔗

Announcing cuda.ml 0.4.0: A major release bringing streamlined NVIDIA GPU-accelerated machine learning directly to R workflows! For data scientists and organizations scaling #RStats analytics, this release delivers faster modeling and seamless GPU hardware integration: Streamlined setup & developer experience: • Two-command installation with automatic GPU backend management • Direct integration with familiar {tidymodels} and {parsnip} syntax • Runs high-performance CUDA workloads without context-switching to Python or managing low-level GPU APIs Expanded tree-ensemble inference: • High-throughput GPU inference for random forests, XGBoost, and LightGBM via nvForest • Portable CPU fallback engine for flexible production deployment • Full serialization support to easily save, bundle, and restore fitted models The release bridges R's ecosystem with NVIDIA's enterprise GPU hardware, making high-performance compute accessible without breaking established analytical workflows. Learn more and try it out: https://lnkd.in/ej8Kwjqu

We have SO many cool things going on next week 👀 Come hangout and learn with us! Tues Sept 1 @ 12pm ET: Join us for a live walkthrough of what's new with the product managers behind Workbench, Positron, Posit Assistant, Posit Package Manager, and Posit Connect. Be there live for the chance to ask questions throughout. Register at pos.it/update-walkthru Tues Sept 1 @ 12pm ET: Learn about Nix and {rix} with the awesome Eric Nantz of R Weekly and the R-Podcast in the cozy, nerdy environment that is the Data Science Lab. Eric will showcase how the {rix} R package (alongside Nix) enables reproducible, isolated, and efficient analytical environments. Register at pos.it/dslab Thurs Sept 3 @ 12pm ET: Rylan Murry, Manager of Data, Analytics & Development at Nebraska Methodist Health System, will be the featured leader at the Data Science Hangout! From being a org's first dedicated data hire to building and running a healthcare data team, Rylan is ready to chat about whatever you want to know. Chat, lurk, or ask questions live! Register at pos.it/dsh As always, you can get a look at the free events coming up on our events page at posit.co/events 👏🏻

Open-source packages made modern data science possible. They also made it a supply chain target. The XZ Utils backdoor (CVE-2024-3094) took two years of patient social engineering to plant. It proved attackers go after maintainers and build chains, not just code. So we rewrote our package security whitepaper for the threat model that exists now. New in the 2026 edition: - Package cooldown: minimum-age rules that let community detection work before new versions reach your users - Compromised maintainer accounts and abandoned packages - VS Code extension governance via Open VSX, the same supply chain risks at IDE-privilege level - The Posit Package Service CDN: why your Package Manager never talks directly to public repositories - Layered blocking: known vulnerabilities (OSV-fed), license rules, and fully custom rules - A decision framework: bottom-up curation vs top-down blocking, and when to use each There is no single silver bullet for the open-source supply chain. Curation, blocking, cooldowns, and unified repositories work together so your analysts keep moving and your security team keeps control. Covers PyPI, CRAN, Bioconductor, and Open VSX. Written to hold up in a security review. Download the updated edition: https://lnkd.in/ebCa8-Me #rstats #python #datascience #infosec #devsecops

Government data science teams shouldn't have to fight procurement to use the open-source tools they already rely on. We've partnered with Carahsoft, naming them Posit's Master Government Aggregator. Federal, state, and local agencies can now acquire Posit's enterprise data science platform through NASA SEWP V and OMNIA Partners contracts they already hold. - No new procurement process: buy through the contract vehicles your agency already uses - A governed, secure environment to build, deploy, and scale with R and Python - Data and code that stay under the agency's control Thank you Craig Abod, Natalie Gregory, and the Carahsoft team. Learn more about the collaboration: https://lnkd.in/efV45i3t

"Can our data scientists use AI assistants?" is quickly becoming "can we prove how they used them?". 🔍 In yesterday's Workflow Demo, Sam Edwardes walked through how Posit Workbench lets admins decide which AI assistants and model providers their teams can use, keep code and credentials inside the managed environment, and produce the audit trail that GRC and Security actually ask for. Check out the recording: https://lnkd.in/gya_weMj

Partnerships get measured in pipeline. They get built on whether the thing you shipped together actually solves something. Our Native App exists because data scientists kept hitting the same wall: their data lives in Snowflake, their work happens in R and Python, and getting between the two meant moving data they weren't supposed to move. Workbench on Snowflake Marketplace closed that gap. The go-to-market work — joint ICP, account mapping, ABM on both sides — is what turned a good product into 173% YoY pipeline growth and 800 joint customers. Thanks to the Snowflake team for the conversation!👇

"Why aren't we nearly as intentional when curating context for ourselves as we are for agents?" - Emily Riederer, Capital One Day 2 of posit::conf(2026) is about what stays human. Emily Riederer (Capital One) Context is the Control Plane. Context mattered to applied data work long before AI made it a buzzword. Emily maps the many kinds data scientists actually need: the unarticulated questions motivating the work, the mechanics of the systems delivering our data, the assumptions baked into every tool, the tacit knowledge sitting in adjacent disciplines. Her argument - curating that context is exactly what keeps this field creative and human. Wes McKinney (Posit, creator of pandas) Navigating the Agentic Future. Agents are taking over a growing share of day-to-day coding work. As models and harnesses improve, where are we still needed? After more than a year of building large new codebases with agents, Wes makes the case that the things that have always mattered - human agency, good taste, architectural thinking - are still what determine whether we succeed or fail. This is the conversation the R and Python community is having about what comes next at posit::conf(2026) September 15 - 16 in Houston, TX. You can be in the room, or join virtually. Register → https://lnkd.in/e7a659Ui