Wes McKinney will be presenting on “The Mythical Agent-Month.”

Abstract: Code generation is converging to free. Agents run in parallel. A single developer can burn 10 billion tokens a month. So why isn’t the software getting better? Drawing on Fred Brooks’s fifty-year-old Mythical Man-Month, this talk argues agentic engineering is repeating software’s oldest mistakes at machine speed: generating technical debt at unprecedented scale while making scope creep unstoppable. Agents can demolish accidental complexity but struggle to distinguish this from essential complexity, and they produce new accidental complexity if not kept in check. The bottleneck was never typing speed. Design taste and knowing when to say “no” were always the hard part. Agents are making this paradoxically even harder.

Find us at Booth #539! We’re bringing a fresh haul of swag, stickers, and a team that’s genuinely excited to hear about your projects and share what we’ve been building.

And stop by on Saturday, May 16 during the afternoon break to grab a copy of Python Polars: The Definitive Guide signed by co-author Jeroen Janssens.

We’re particuarly excited to chat about:

Quarto

Open-source scientific and technical publishing system built on Pandoc

5.3k

Best Practices Publishing

Positron

Positron, a next-generation data science IDE

4k

Best Practices Interactive Apps

Great Tables

Make awesome display tables using Python

2.6k

Best Practices Publishing Visualization

George Stagg will present on “Effective Agents: A Builder’s Guide to Working with AI”.

Description: The gap between a working demo and a reliable agent is where most of the interesting engineering happens. In this talk, we’ll build up a practical mental model for working with AI. I’ll start with some perhaps surprising fundamentals, and then share some of our hard-won lessons from shipping agents in production.

LLM APIs behave differently from other services you might have worked with before. They’re stateless at the network level, context is everything, and a lot of surprising behaviour only becomes obvious when you look at the raw traffic. They’ll also lie to you: confidently reporting that a capability is working when it isn’t, or finding creative workarounds to limitations you didn’t intend to be optional. So, I’ll go over the anatomy of an AI network request and show why proxying your own traffic is one of the most useful tools you can develop in this space.

From there, we’ll move into the practical realities of building and deploying agents: how to think about context management, why caching deserves more attention than it often gets, and when abstractions help versus when they obscure what’s actually happening. I’ll share what we learned from agentic tools like Claude Code, covering the best way for agents to interact with the world, why multi-agent systems with limited scope and focused toolsets outperform monolithic agents, and what increasing autonomy and abstraction means for how we build and debug. I’ll end with what we’d do differently if we were starting over.

Neal Richardson will present on " MCP, or not MCP".

Description: Model Context Protocol is a standard for defining tools that can be made available to LLMs and AI applications. There’s a lot of noise out there about what you should use to get the best results from AI, so in this talk, I will provide some guidance on when you should use MCP, and when you should reach for some other tool. I will describe cases where MCP is the right tool for the job, and when other things, like skills or other context files, are better. I will also devote attention to questions of security and authentication, which are important for MCP, and provide concrete examples of how MCP servers can be used to unlock agentic workflows while also strengthening data governance. This talk is intended for those who are interested in using LLMs for workflows involving data. No prior experience with MCP is required.

Neal Richardson will present on “MCP, or not MCP”.

Description: Model Context Protocol is a standard for defining tools that can be made available to LLMs and AI applications. There’s a lot of noise out there about what you should use to get the best results from AI, so in this talk, I will provide some guidance on when you should use MCP, and when you should reach for some other tool. I will describe cases where MCP is the right tool for the job, and when other things, like skills or other context files, are better. I will also devote attention to questions of security and authentication, which are important for MCP, and provide concrete examples of how MCP servers can be used to unlock agentic workflows while also strengthening data governance. This talk is intended for those who are interested in using LLMs for workflows involving data. No prior experience with MCP is required.

Christophe Dervieux and Maëlle Salmon will lead a workshop, “PDF sans frictions : Typst dans vos projets Quarto”.

Description: Ce tutoriel présente comment produire des PDF de qualité avec Quarto en tirant parti du format Typst. Après un bref rappel des bases de Quarto, un focus sera consacré à Typst : avantages, cas d’usage et différences avec les chaînes PDF classiques. L’usage de _brand.yml sera présenté pour centraliser les logos, les palettes de couleurs et les choix typographiques afin d’obtenir des rapports cohérents avec l’identité d’une organisation. Le rôle des templates sera expliqué : adapter et réutiliser des mises en page pour répondre aux besoins d’un laboratoire, d’une équipe ou d’un projet.

Une partie pratique permettra de mettre en place un petit projet Quarto + Typst, d’appliquer un _brand.yml, et de créer ou modifier un template. Enfin, une introduction aux extensions et aux bonnes pratiques montrera comment pérenniser et partager les personnalisations.

Kristin Bott will present on “ellmer for all? : building context around LLMs”.

Description: Large language models (LLMs) are changing the way people write code and build software; LLMs are also increasingly present in the work lives of people who don’t write code. How do you connect with a diverse audience to make them effective users of LLMs, and more comfortable having technical conversations around these tools? In this talk I will share some lessons learned from initiatives within Posit to build increased end-user comfort and key knowledge around LLMs, both for folks who use AI daily and for people who have never (yet) seen ellmer in action.

Charlotte Wickham will lead a workshop, “Zero to Website with Quarto Workshop”.

Description: As a data professional, you’ll eventually want a website—whether for yourself, a project, or a research group. Quarto is an excellent way to build one: create your site with plain text files that you can edit in your favorite tool and check into version control; organize your content into folders and pages in a way that makes sense to you; create content that includes code and its output; and generate listing pages that help readers navigate collections of content like projects, publications, or people. In this workshop, you’ll walk through the process from scratch. You’ll see how to get started with a basic template and then customize it to your needs, including adding pages, customizing navigation, adding a blog or other listing, personalizing the appearance, and getting it online.

This workshop will be hands-on, and you’ll need to prepare a few things before it begins. Look for instructions closer to the workshop date. It will help if you’ve already used Quarto to produce documents, but I’ll provide resources to get you up to speed beforehand if you haven’t. You don’t need any HTML, CSS/SCSS, or Git/GitHub experience, nor do you need any particular programming language (R, Python, etc.) or level of programming experience.

Charlie Gao will present on “CRDTs for R: Conflict-Free Data Structures for Real-Time Collaboration”.

Description:

Collaborative data analysis presents a fundamental concurrency problem: when multiple users modify the same data simultaneously, how should conflicts be resolved? Traditional approaches rely on locking or central arbitration, but conflict-free replicated data types (CRDTs) offer a principled alternative. A CRDT is a data structure whose concurrent operations are guaranteed to converge to the same state, regardless of the order in which they are applied. This mathematical property — strong eventual consistency — eliminates the need for conflict resolution logic entirely.

We introduce CRDTs to the R ecosystem through automerge, an open-source package developed at Posit that natively bridges R’s data model with the Automerge Rust CRDT engine. Collaborative documents — maps, lists, and text — appear as familiar R objects, but every edit is tracked as an individual operation under the hood. When documents are modified concurrently, changes merge automatically.

Beyond real-time document editing, CRDTs open up new possibilities for the R ecosystem: shared annotation of datasets across a research team, collaborative model specification, or any workflow where multiple analysts need to work on the same objects without coordination overhead. We make this concrete with a live demonstration: audience members will open a Shiny application on their own devices and edit shared state together, watching changes from every participant merge seamlessly in real time.

Tomasz Kalinowski will present on mcp-repl.

Description: R is an interactive environment. Working effectively with R means being able to interact with a live session to inspect objects, view and iterate on plots, access help, and step through running code in the debugger. Making LLMs effective in R therefore means more than giving them a way to execute code: it means exposing R’s interactive affordances in a form the model can use.

This talk introduces mcp-repl, an open-source front end to R that plugs into existing agents such as Claude and Codex. It provides a long-lived REPL with persistent state, in-band help, plot capture, support for debugger- and readline-based workflows, guardrails for large outputs, responsive handling of both interactive and long-running computations, and OS-level sandboxing. In practice, mcp-repl makes general-purpose coding agents much more capable at package development, Shiny debugging, and iterative analysis, especially in long-running autonomous tasks without a human in the loop.

Because mcp-repl plugs into existing agents, it also reflects the constraints of a plug-in architecture. To make that contrast concrete, this talk also looks at Posit Assistant as an example of what becomes possible when the interface is integrated into an IDE for data analysis and designed to keep the human in the loop.

Together, Posit Assistant and mcp-repl illustrate two ways to bring R’s interactive affordances to an LLM: one through an integrated experience designed for close human collaboration, the other through a plug-in front end designed for autonomous work in a private runtime.

We are proud to be sponsoring this year’s event to help support the incredible work happening in the scientific Python ecosystem. We’ll have a dedicated space for demos and deep dives, and will share more specific information on our booth location as soon as it is available.

We’ll be updating this page with more details, including booth locations, talk schedules, and potential swag, as we get closer to the date.

See you in Amsterdam!

Simon Couch will present a keynote on lessons from building competent data science agents.

Nick Strayer from Posit presented on “Positron: Data Science Without Constraints”.

Description: There’s no shortage of explanations of what Positron does, but the why behind those features is just as important. Positron builds on lessons from RStudio and other data science IDEs, with the goal of supporting exploratory, collaborative, and multi-language data science work without switching tools. Loved your workflow in RStudio but suddenly found yourself on a Python-based team? Positron lets you bring your favorite tools, like the plots viewer and variables pane, into a Python-based workflow. That same philosophy extends to how you work with notebooks. Positron’s new notebook editor is designed to work across languages and notebook formats while taking full advantage of the surrounding IDE, bringing tools like the data explorer and Positron Assistant right into your notebook workflow. We’ll end by highlighting what’s new and upcoming in Positron as it continues to pursue this vision.

Charlotte Wickham led a workshop on “Building Accessible, On-Brand Documents with Quarto”.

Description: Come see practical strategies for producing Quarto documents that meet organizational standards for both design and accessibility. You’ll learn how to implement consistent organizational branding using brand.yml, plus customization techniques for cases where you need more control. You’ll also learn about recent accessibility improvements for both PDF and HTML outputs.

François Michonneau led a workshop on “Working with Larger than Memory Data in R”.

Description: As datasets continue to grow in size and complexity, R users increasingly encounter data that exceeds their system’s memory capacity. This hands-on workshop provides practical strategies for efficiently analyzing larger-than-memory datasets using modern open source tools, with a focus on DuckDB and Apache Arrow—all while maintaining familiar tidyverse workflows.

Participants will learn when and why to move beyond traditional in-memory data frames, and how to choose the right tool for their specific data challenges. Through a combination of presentation and hands-on exercises, we’ll explore how DuckDB enables SQL-based analytics on large datasets without loading them entirely into memory, and how Arrow provides a high-performance columnar data format for efficient data interchange and processing. We’ll also introduce duckplyr, which brings DuckDB’s performance optimizations directly to your existing dplyr code with minimal syntax changes.

The workshop covers essential workflows including reading and querying large CSV and Parquet files, performing aggregations and joins on data that won’t fit in RAM, and leveraging duckplyr to accelerate familiar tidyverse operations on larger datasets. Participants will gain practical experience through real-world examples and learn decision frameworks for selecting appropriate tools based on data size, query patterns, and performance requirements—all without abandoning the tidyverse syntax they already know.

Sara Altman led a workshop on “LLMs for Data Analysis in R”.

Description: LLMs are transforming how we write code, build tools, and analyze data. This workshop will introduce participants to programming with LLM APIs in R using ellmer, an open-source package that makes it easy to work with LLMs from R. We’ll cover the basics of calling LLMs from R, system prompt design, tool calling, and evaluation, and show how to use LLM-powered tools to support common data analysis tasks like exploratory data analysis. Participants will leave with example scripts they can adapt to their own data analysis projects.

Mine Çetinkaya-Rundel presented a short course, Getting Started with Positron: A Next-Generation IDE for Data Science.

Description:

Positron is a next-generation data science IDE built by Posit PBC that combines the best features of RStudio and Visual Studio Code. This tutorial will introduce Python and R users to Positron’s core capabilities, with an emphasis on helping RStudio and VS Code users while highlighting its seamless integration with the respective language ecosystems.

For R programmers coming from RStudio, Positron delivers a familiar yet enhanced environment for data analysis and package development, while offering a path to Python when needed. For Python programmers coming from VS Code, Positron offers data science–specific enhanced capabilities for exploring and analyzing data, authoring, and publishing. Unlike traditional software development–oriented IDEs, Positron provides first-class support for data science–specific workflows through its native support for R and Python, along with a designated, integrated UI for exploring for variables (environment), connections, plots, help, and more.

Caleb Kinney presented on “Why Vulnerability MTTR Alone Misleads: Add MOVA to Measure Real Risk”.

Description: Teams celebrate when their Mean Time to Remediate (MTTR) drops until it suddenly spikes after fixing old vulnerabilities. That looks like failure, but it’s actually progress and exposure went down. MTTR measures how quickly work closes, not the health of what remains open. Mean Open Vulnerability Age (MOVA) fills that gap by showing the average age of open vulnerabilities at a given point in time, revealing true backlog risk.

This talk defines MTTR and MOVA in clear, practical terms and walks through a simple simulation comparing two common fix strategies: newest-first and oldest-first. MOVA brings that missing dimension by translating backlog health into data leaders can act on. Attendees will see why MTTR alone can mislead, how MOVA exposes hidden risk, and how combining both metrics gives security teams and leaders a more accurate picture of progress and exposure.

Rodrigo Silva Ferreira from Posit presented “To Notebook or Not to Notebook: Multilingual Workflows for Exploratory Data Analysis”.

Description: Notebooks have become the default environments for exploring data, prototyping ideas, and sharing results. Yet, their influence extends far beyond code execution, as they shape how we think, collaborate, and communicate with data. As real-world projects increasingly rely on multiple programming languages, magic commands and cross-language kernels open new possibilities for combining Python, R, SQL, markdown, and more within a single workflow, blending computation with storytelling in powerful ways.

This talk explores how multilingual workflows transform exploratory data analysis (EDA), highlighting where notebooks shine and where they introduce new challenges around testing, version control, and scaling. Through real-world examples, we’ll look at how these tools lower barriers, reinforce trust and integrity in data science workflows, and change the way we reason about problems. Attendees will gain practical strategies for integrating multiple languages in their notebooks and a deeper understanding of how these tools shape modern data science. Basic familiarity with Python and EDA is helpful but not required.

The conference features a tutorial day on Monday, December 8, followed by two days of talks, keynote presentations, lightning talks, and networking opportunities on December 9-10. PyData Boston provides an excellent opportunity for attendees to learn about the latest developments in the PyData ecosystem, share knowledge, and connect with peers from around the world.

Keynote speakers include Lisa Amini from IBM and Isabel Zimmerman from Posit, who will share insights on the future of data science and development tools. The event is supported by major sponsors including Microsoft (host), Anaconda, NVIDIA, Snowflake, and Posit, demonstrating strong industry backing for the Python data community.

PyData conferences are community-driven events that foster collaboration and knowledge-sharing within the open-source data science ecosystem. Whether you’re working on machine learning, data visualization, statistical analysis, or building data infrastructure, PyData Boston offers valuable sessions and connections to advance your work.

Hosted in the centre of Newcastle Upon Tyne, UK, this conference delves into the world of {shiny} and other web-focused R packages.

Charlie Gao, Senior Software Engineer on Posit’s open source team, reviewed some of the latest high-performance async tooling developed by Posit to support R Shiny in terms of performance, scalability and user experience.