So Much to See, So Little Time: Your posit::conf(2026) Agenda Breakdown

Table of Contents

• AI in Practice: Moving Past "Should We?" to "How Do We Do This Well?"
• What Does Literate Programming Actually Look Like in 2026?
• Quarto as Collaboration Layer
• SQL: Still Here, Still Doing Great
• Building Shiny Apps Like You Mean It
• Taking Your Work to Scale
• R and Python at the Point of Care
• Keynotes Worth Showing Up For
• A Few More Sessions You Shouldn't Miss
• Don't Overlook the Workshops
• Register Before Spots Fill

The conference for data scientists, R and Python practitioners, and open-source builders comes to Houston this September. Here's a deep look at what's coming.

If your data science workflow looks different today than it did a year ago, you're not imagining it. The tools we use, the way we think about notebooks, how we collaborate, how we work alongside AI — all of it is shifting, sometimes faster than we can keep up with. posit::conf(2026) is where our community comes together to make sense of it.

Running September 14–16 in Houston, the conference kicks off on September 14 with a full day of hands-on workshops (in-person and virtual), followed by two days of 119 talks across 30 sessions. Whether you're navigating agentic AI in your codebase, scaling a Shiny app to production, or just trying to figure out what literate programming should look like in 2026 — there's a room for you.

Here's a look at some of the threads we're most excited about.

AI in Practice: Moving Past "Should We?" to "How Do We Do This Well?"

Last year, a lot of the AI conversation at data science conferences was about possibility. This year, the posit::conf program is squarely focused on practice — with all the friction, failure modes, and hard-won design decisions that come with it. If you've been using AI tools in your work and wondering whether you're doing it right, these sessions are for you.

"Being an Informed and Savvy AI User" (Sep 15) is a good place to start if you want to be a more critical, intentional AI user. Allissa Dillman and Padmashri Saravanan kick things off with a talk on how LLMs embedded in our IDEs don't respond neutrally — the framing of your question, and even implicit cues about who's asking, shifts what the model produces. You'll leave with a sharper instinct for when to trust AI-generated code and when to push back.

The session also features J.J. Allaire introducing Inspect Scout, Posit's new Python framework for systematically evaluating and monitoring AI agents through transcript analysis. If you've ever shipped something agentic and couldn't explain why it failed, this talk is directly aimed at your problem. You'll walk away knowing how to build evaluation into your agent workflows from the start — not as an afterthought.

"Agents, Context, MCP" (Sep 15) goes deeper into the infrastructure layer. Neal Richardson's "MCP, or Not MCP" unpacks Model Context Protocol — the emerging standard for exposing tools to LLMs — and gives you a clear-eyed view of where it genuinely helps and where the abstraction leaks. Sharon Wang's "What Context Do Your AI Agents Actually Need?" tackles one of the most underappreciated questions in agent design: not what you can put in a context window, but what you should, and what happens to performance when you get it wrong. These are the kinds of calibration questions that will save you a lot of debugging time.

James Wade's "AI Agents Deserve R" is the talk many long-time R practitioners have been waiting for. If you've been feeling pressure to rewrite everything in Python to work with agentic frameworks, Wade makes a compelling case for staying in R — not out of nostalgia, but because ellmer, shinychat, and ragnar have matured into production-viable building blocks. You'll walk away with a clearer picture of where the R ecosystem actually stands for agent development today.

"AI in Data Science" (Sep 16) is the most practically skeptical session in the program — and all the better for it. Max Kuhn's "Roll for Wisdom: What AI Gets Wrong About Data Analysis" is a methodologically grounded look at how LLM-generated analysis fails: the mistakes that look right until they don't, and the structural reasons they happen. Nicky Bell follows with "Between Code and Conclusions: An Evaluation Layer for LLM-Generated Analysis" — a systematic approach to auditing AI-produced insights before they make it into a report or a decision. You'll come away with a concrete framework for knowing when to trust AI-assisted analysis and when to dig deeper.

The throughline across all three sessions: the community is shifting from "how do I use AI?" to "how do I know when AI is wrong?" — and that's exactly the right question to be sitting with right now.

What Does Literate Programming Actually Look Like in 2026?

This is a question the community has been debating for years — and "The Notebook" (Sep 15) takes it head on with four speakers bringing genuinely different perspectives. Whether you're a devoted Quarto user, a Jupyter loyalist, or somewhere in between, you'll leave with a clearer sense of where the tooling is headed and what trade-offs you're actually making.

Jonathan McPherson introduces Quarto Notebooks in Positron — a new feature that gives you access to Jupyter's literate programming experience while keeping the plain-text source file that makes version control, code review, and collaboration tractable. It's a thoughtful architectural bet, and McPherson explains the reasoning behind it. You'll understand why the notebook UI and the plain-text format don't have to be at odds.

Nick Strayer's companion talk, "Positron Notebooks: Built Around How You Actually Work," is grounded in something we don't see often enough in tool design: real user research. The Posit team talked to data scientists across many kinds of teams about their actual day-to-day notebook habits — not the idealized version, but what people actually do under deadline pressure with messy data. Hearing how those conversations shaped the design will give you a new appreciation for the choices Positron makes.

Tom Shafer brings a fresh perspective with Marimo, a modern Python notebook system built on different principles: no hidden state, no out-of-order execution, a storage format that's actually diff-friendly. If you've ever pushed a Jupyter notebook to git and watched the merge conflict horror unfold, this talk will feel personally relevant — and you'll come away with a concrete alternative worth trying.

Paula Lopez Casado (ex-Apple) rounds things out with the craft side of the equation: how to elevate Quarto reports through design, layout, and narrative structure so they actually influence decisions rather than just document them. A practical, immediately applicable session for anyone whose reports need to land with non-technical stakeholders.

Quarto as Collaboration Layer

Quarto has moved well beyond "a nicer R Markdown." It's becoming connective tissue for the data science organization — and several sessions this year explore what that means in practice.

The headline in "Quarto, Connected" (Sep 16) is Quarto Hub, introduced by Carlos Scheidegger: an open-source and hosted platform for collaboratively creating, editing, commenting on, and publishing Quarto projects. Think Google Docs-style collaboration, but for reproducible documents with embedded code. If your team has been passing .qmd files back and forth over email or struggling with async editing, this is worth paying close attention to.

Emil Hvitfeldt makes the case in "Slideware Is Dead — Long Live Quarto" for ditching presentation tools entirely in favor of Quarto reveal.js presentations. Not just for reproducibility reasons, but because the constraints of slide-native tools have been quietly distorting how we communicate technical work. You'll come away either convinced or ready to argue — and either way, thinking more deliberately about your presentation choices.

Charlotte Wickham closes the session with a focused, practical look at accessibility in Quarto: what it actually takes to produce accessible output, not just output that renders without errors. This is the kind of talk that changes your default habits. And on the virtual day, Björn Fisseler's "Shift-Left for Accessibility" gives you the workflow-level tools to prevent accessibility debt from building up in the first place.

SQL: Still Here, Still Doing Great

If you were expecting the data stack session to relitigate whether SQL still matters — good news, it doesn't bother. "The Modern Data Science Stack... Still SQL" (Sep 15) just gets on with making SQL better, and there's a lot to be excited about.

Julia Silge and Brian Lambert open by making the case for SQL as a first-class language in your data science IDE, with Positron tooling to match. Then Thomas Lin Pedersen introduces ggsql — a grammar of graphics for SQL. If you spend significant time in database-native environments, the ability to build visualizations directly in SQL is a genuine workflow unlock. You'll see what that looks like in practice.

Ian Cook and Bryce Mecum present ADBC (Arrow Database Connectivity) — a faster foundation for database connectivity that replaces the ODBC/JDBC bottleneck with Arrow-native data transfer. The performance implications for large-result queries are significant, and you'll understand exactly why. Brendan Broderick closes with "SQL Without Writing SQL: dbplyr for Serious Shiny Apps" — a session that reframes dbplyr not just as a convenience layer but as an architectural pattern for building Shiny apps that stay performant as data scales. A great one if you've ever had a Shiny app slow to a crawl on a growing dataset.

Building Shiny Apps Like You Mean It

The Shiny sessions at posit::conf(2026) reflect a framework that has genuinely grown up. This isn't small-team dashboard territory anymore — these talks are about production software with real engineering requirements, and the community building it has the scars to prove it.

"Shiny in Pieces" (Sep 15) focuses on architecture. Barret Schloerke's "Beyond Bootstrap: Building Custom Shiny UI Packages with AI" is a practical demonstration of using AI to scaffold bindings to modern React component libraries — Material UI, Ant Design, and others that have previously been out of reach for most Shiny developers. You'll come away knowing how to build Shiny apps that look and behave like first-class web applications, not BI tools.

Isabel Glauss presents the DaVinci framework from Boehringer Ingelheim — a modular Shiny system built for clinical trial data exploration that directly addresses the triad of code duplication, UX inconsistency, and validation expense in regulated environments. If you're building Shiny in pharma or any regulated industry, this is directly applicable and worth taking notes on.

Karan Gathani's "Quality Testing in the Age of Excess" asks a question that's newly relevant: when AI code generation makes it trivially cheap to write tests, what does rigor actually mean? The economics of software quality have shifted, and you'll walk away with a sharper view of what "well-tested" should mean in an AI-assisted development workflow.

Taking Your Work to Scale

"From Laptop to Cluster" (Sep 16) is for anyone who's outgrown local computation and needs a principled path forward — without rewriting everything from scratch.

Charlie Gao's "Stop Waiting for Parallel R" introduces async parallelism in R: a fundamentally different model from the batch-and-wait approach most of us are used to. Async means you're not locked out of your session while tasks run. You'll leave understanding what async parallelism actually enables and how to start applying it.

Travis Gerke's "Quack to the Future: Building Traceable Data Lakehouses with DuckLake in R" brings together two exciting developments — DuckDB's continued expansion and the data lakehouse pattern — into a practical R workflow with built-in data lineage. You'll see a concrete architecture you can adapt. Edgar Ruiz closes with a demonstration of scaling Tidymodels on Databricks with sparklyr: prepare locally, run at scale, keep the tidy API. A clean, practical pattern for teams at the threshold of needing distributed compute.

R and Python at the Point of Care

posit::conf has always had a strong clinical and public health track, and 2026 is no exception. The "Clinical Data Science" session (Sep 16) is a reminder of what's possible when data science is applied with real care to real-world health outcomes.

You'll hear how Neil Rosenbaum's team at Houston Methodist built a Posit/R-powered pipeline to quantify how environmental stressors — heat events, air quality — drive emergency department utilization, enabling proactive capacity planning before strain hits. Rebecca Wardrop shows how Quarto-generated performance reports at Michigan Medicine have been used to drive down cesarean rates through data-informed clinical quality improvement — reproducible reporting with measurable impact. Florian Mayr walks through building an ICU dashboard at the VA with Shiny, the kind of production data product where the stakes of getting it right are genuinely high. These talks are a powerful reminder of why we do this work.

Keynotes Worth Showing Up For

Four keynote speakers anchor the conference across both in-person days.

Sara Altman and Simon Couch open the conference on September 15. Both are on Posit's AI Core team, where they spend their days building data science agents and rigorously evaluating how well those agents actually handle analytical work. Their talk cuts through the AI noise to address a specific and underappreciated problem: while good data analysis requires slowing down, questioning conclusions, and sitting with uncertainty, most of today's AI agents are wired to keep moving. Expect practical strategies for keeping AI-assisted analyses correct, transparent, and reproducible — from people who are building this stuff, not just theorizing about it.

Christine Zhang, Graphics Editor at The New York Times, closes Day 1. Zhang operates in one of the highest-stakes data visualization environments in the world — building pipelines that have to hold up under breaking news conditions and public scrutiny simultaneously. Her session focuses on resilient workflows, the discipline of quantifying and communicating uncertainty, and what user-centric design actually means when your audience has no statistical training. (She's also a longtime R user — she's in the credits of the readxl 1.1.0 release notes, and uses summarise() and colour() — so she'll feel right at home.)

Emily Riederer from Capital One opens Day 2. Riederer has worked up and down the data stack, and her talk draws on that range — exploring the "big ideas" embedded in different data disciplines and how to apply engineering principles to build more resilient, developer-friendly data products. Expect a full-stack mindset: what best practices hold universally, which ones are context-dependent, and how polyglot experience across tools exposes modes of thinking that transcend any single language.

Wes McKinney closes the conference. As Principal Architect at Posit — and the creator of pandas and a driving force behind Apache Arrow — McKinney is now focused on what he calls the "Mythical Agent-Month": why adding more AI agents to a software project doesn't automatically make it better or faster. His session moves past the hype to address the real productivity paradox of agentic development, and what it means to design software that is genuinely agent-friendly rather than just AI-adjacent.

A Few More Sessions You Shouldn't Miss

A few threads in the program that deserve mention:

The Lightning Talks session on Sep 15  includes Daniel Falbel on raghilda — Posit's approach to Retrieval-Augmented Generation in Python — and Richard Iannone on Great Docs, a toolkit for building beautiful documentation sites for Python packages. Lightning talks are always a high-density signal source.

‘Taming Messes with AI’ (Sep 16) is a data wrangling session dressed up in LLM clothing — and it's the better for the combination. Leslie Emery's talk on configurable clinical data harmonization using LLMs addresses one of the most stubborn problems in life sciences data work. Mirian Lima from the United Nations presents a context engineering approach to policy analysis that makes large-scale document processing tractable on a constrained budget.

‘R Nerds’ (Sep 16) is exactly what it sounds like: Sierra Johnson on R optimization (caching, compiling, the five stages of grief), Meghan Harris on mocking for testing in production R environments, and Jon Harmon on the design principles behind well-crafted R function APIs. If you care about writing R that other people can actually use and maintain, this session is essential.

This is also one of the things that makes posit::conf different: the hallway conversations, the Birds of a Feather sessions, the chance to finally meet the person whose package you've been using for three years. The formal program is only part of what you'll take home.

Don't Overlook the Workshops

Before the main conference begins, September 14 is a full day of hands-on workshops — and the instructor lineup alone is worth the trip. In-person workshops run 9am–5pm in Houston. There's also a virtual half-day workshop repeated across three time slots to cover every major timezone.

This year's offerings span the full range of the Posit ecosystem: Hadley Wickham and Jenny Bryan on the modern R workflow (featuring AI and Positron), Max Kuhn and Emil Hvitfeldt on practical machine learning with tidymodels, Charlotte Wickham and Mine Çetinkaya-Rundel on advanced Quarto patterns and tooling, Garrick Aden-Buie and Sara Altman on programming with LLMs in R and Python, and more — including a deep dive on orchestrating AI-ready statistical infrastructure for regulated environments and a Pharmaverse clinical reporting workshop for those working in regulated environments. The virtual option covers getting started with Positron for both R and Python users. And if you're not in a workshop, September 14 also features a full slate of virtual-only talks — three dedicated sessions covering general data science topics and modelling — so there's no such thing as a slow start to the week.

Workshops are separate from general registration and fill up fast — often before the main conference sells out.

See the full workshop lineup →

Register Before Spots Fill

posit::conf(2026) runs September 14–16 in Houston, TX, with a virtual day on September 14 for remote attendees. Whether you're coming for the AI infrastructure content, the Quarto and Positron updates, the clinical data science track, or just to be in the same room as the practitioners building the tools you use every day — this is the conference worth making space for.

Register for posit::conf(2026) →

See you in Houston and online!