Transcript#

This transcript was generated automatically and may contain errors.

Hi everyone, and welcome. We're really glad you're here. I'm Nick Rohrbaugh, and I lead product marketing for Positron and Posit AI, and I'll be your moderator for the next hour. Today is about two new capabilities in RStudio and Positron. Posit Assistant, our new data science agent, and NextEdit Suggestions, which are AI-powered code suggestions that appear as you type, and Posit AI, the new subscription that powers it all. You'll hear from four people on our product and engineering teams, Sarah, Simon, Alex, and Tom.

A few quick logistics, and then we'll get started. We're currently live streaming to YouTube and on Zoom, and we're recording. You can find the full recording here on YouTube after the event. If you're joining us on Zoom, you can drop questions in the Q&A panel anytime, and we'll save some time at the end to answer them. Also on Zoom, the chat is open if you want to say hi or connect with each other. All right, over to you, Sarah.

Why data scientists need specialized coding agents

Over the last year or so, the coding world has really changed. Code is increasingly written by or with the help of coding agents. And working with data involves writing code, but it isn't really the same as pure software engineering. Data scientists, researchers, analysts, et cetera, have particular needs and constraints because your job is not just to build tools, but also to develop an accurate understanding of the world through your data.

And because of this, data scientists need coding agents built for them. At Posit, we've been making coding assistants or agents for people who work with data since about 2024, which is really as long as this was even a feasible option. And today, we're going to talk about our latest, most advanced and mature agent, Posit Assistant. Posit Assistant, as you'll see in a bit, is both a general purpose coding agent and a specialized data analysis agent. And it lives right in RStudio and Positron.

Evolution from DataBot to Posit Assistant

If you've been following Posit's agent development, you might have noticed an evolution in our thought process about agents for data science. We've gone from caution and even some pessimism to where we are today, which is confident in the usefulness of these tools and in our ability to make them. So in August of 2025, we released DataBot, which is an experimental exploratory data analysis agent available in Positron. And we were really excited, but also worried about the risks of DataBot incorrectly analyzing data. And so DataBot was purposely constrained just to EDA and also released wrapped under several layers of warning. We even had a literal caution sign as the image of one of our blog posts.

So we went from releasing DataBot in August of 2025 like this to where we are today with Posit Assistant and the Posit AI service, which I think is a much more confident, optimistic place. So what changed? The first thing is that the models really have improved dramatically. This makes the possibility of catastrophic mistakes generally go down and also just means that the models are much more useful as agents today. Model improvement is sort of like the rising tide that lifts all the boats of agent behavior. We've also taken everything that we've learned from our past experiments, including with Positron Assistant and DataBot, as well as what we find useful with general purpose coding agents like Cloud Code. We've taken all of that learning and put it into Posit Assistant.

And finally, I think our perspective on risk has shifted slightly. We're definitely still worried and careful about models making mistakes with real world consequences, but there is risk on the other side as well. At this point, these tools are really too useful to ignore and you, our users, do important work and you deserve access to the state of the art.

At this point, these tools are really too useful to ignore and you, our users, do important work and you deserve access to the state of the art.

Demo: Posit Assistant in RStudio

Okay, so now we're going to go into some brief demos of Posit Assistant and NextEdit suggestions in RStudio and Positron. Okay, so Posit Assistant lives in the sidebar of RStudio and it can basically help you with any coding or data task that you have. So let's ask it to analyze some data. I have these two ocean-related CSVs over here. They have ocean temperature in Nova Scotia at various depths. Let's ask it to read in one of those and start analyzing it. You can use the app to reference specific files.

Okay, so Posit Assistant went ahead and it read in that data set. It's giving us a description of the columns and their types. And then it is giving us some suggestions here. So next, let's take one of its suggestions and check for missing values.

So Posit Assistant has built-in access to your R session. It can automatically see any objects that you have loaded. It can run code. It can access plots. It can see your console, all of that, and this is all built in automatically. Okay, so it has run some code to start looking at missing values in the data. And one thing that you might notice is that when Posit Assistant is analyzing data, it has this pattern where it runs some code, it gives you a summary, maybe with a table like you can see there, and then it offers some suggestions for where to go next. So let's do one more thing. Let's make a visualization. We'll visualize temperature over time.

Okay, so it is making a plot of ocean temperature over time. You can see the seasonality trend here. Great. Okay, so so far we've used Posit Assistant to analyze the data. We read it in. We made a plot. You can also use Posit Assistant for more coding-focused tasks. One thing you might do here is make a Shiny app. You can also use Posit Assistant for things like building R packages, writing a series of scripts, all of that, any general-purpose coding task in R or Python. For now, let's ask it to make a Shiny app to display this data.

Okay, you'll notice that it loaded the Shiny BSLib skill. Skills are additional pieces of prompting and maybe some code that give Posit Assistant access or more information about specific abilities. This skill just provides more information about how to make a Shiny app with BSLib.

Okay, so it's asking for permission to write an app.r file. Let's allow it. And now it's going to do a Shiny app. Let's allow it to do a Shiny app. And now it's going to add the app code to app.r. And here is the app. Great.

In real life, we might, you know, keep iterating on this app with Posit Assistant. For now, that looks pretty good. And it's just giving us a summary of what it did.

Okay, so the next thing I want to do is add the analysis we've done so far to a report. You can do this with the report command. This is going to create a Quarto document with our analysis that we've done so far in the chat. I'm just going to give it a brief description of what to do. You can be more specific with it if you want. Okay, and it again is loading a skill. It's loading the report skill, which has instructions for creating Quarto documents or Jupyter notebooks. And one of the things that skill instructs it to do is propose an outline for your approval. So, let's just say this looks good.

And now it is going to add our analysis into the Quarto report.

While it does that, one thing I want to point out is we have thinking turned on. So, you see that little thinking text, and now we can expand it and see the actual thinking text. And this is the model working through a problem before it's giving you its final answer. You can turn thinking on or adjust the level with this icon down here, the label icon. So, you can turn it off or adjust the level.

Okay. So, now our Quarto report is ready. At this point, you can edit the Quarto report, ask Posit Assistant to make changes, and continue iterating. I want to point out plan mode, which we can enable with this button here. And this lets you and Posit Assistant create a plan for your task before Posit Assistant actually starts executing it. We can also select the model with the model picker. This allows you to pick the model that best suits your use case.

Demo: Posit Assistant in Positron

Okay. Now let's take a look at Positron. So, Posit Assistant is now available in Positron, and it works much the same way as we just saw in RStudio. It has access automatically to your R or Python session. So, let's take a look at Posit Assistant in Python. Right now I have an R console going. And so, you can see the little R in Posit Assistant. Let's switch to Python. I'm just going to click on the Python console. And now it says Python down here. So, let's ask it to do the same thing that we did before and read in the ocean temperature data.

Okay. So, it's using pollers now to read in the data. It's giving us, again, a bit of a summary, and then offering us suggestions for where to go next.

Demo: NextEdit Suggestions

Thanks, Sarah. That's Posit Assistant. The other thing Posit AI powers is NextEditSuggestions. Simon, walk us through it. So, Sarah just demoed Posit Assistant, and Posit Assistant is sort of one of two products that currently ships with the Posit AI service. The other one is NextEditSuggestions.

Posit Assistant generated a little report, and I just had Sarah send me that report over. And so, I have it pulled up inside of my editor now. And I sort of stubbed it out, so we just have those first few paragraphs and lines of code. And so, all I've done up to this point is load in that ocean temperature data. So, NextEditSuggestions is most useful when you're working in the editor already. If you've used IDEs like VS Code or Cursor, they each have these NextEditSuggestions or SuperComplete or SmartAutoComplete, whatever you want to call them, features, where as you type, there's sort of smart and reasonable suggestions about what you might type next.

So, for example, I might type something like ggplot, and before I can think of what to type next, ocean temp, which is the correct name for that variable, has been filled in here. So, this is something you could probably expect from those other NextEditSuggestions systems, where it can read the lines above your cursor. So, right here on line 28, I've already typed ocean temp. So, just from the lines surrounding that line, the model can tell that there's a variable called ocean temp in my environment.

Where NextEditSuggestions from Posit.ai will start to differ from your experience with those other tools is in its own ability to communicate with the active R or Python session. So, if you've worked with those other systems before, you might anticipate that once you type a plus, there will be column names that are hallucinated, because whatever that system is doesn't actually have access to the names of the columns inside of your workspace. But if I type here, and I'll say AES, you can see that the completions here are suggesting date and mean temperature degree Celsius. Those are both actual column names inside of the data ocean temp. So, if I tab to accept that, and then maybe also add a G online, we can see that the plot it's come together with here actually executes.

The other thing that you might have noticed is that I'm inside of a Quarto document right now, and the NextEditSuggestions system is not messing that Quarto document up. So, if you haven't seen or tried to work with one of these systems before, that might sound like a given, but this is actually something that many of those other systems really struggle with, and we worked hard to get this right. So, as you work inside of Quarto documents, that system is not going to mess up the formatting of the code cells or whatever it may be.

Posit AI subscription

Thanks, Simon. So, that's the experience, the Posit Assistant agent and the inline completions from NextEditSuggestions. Next up, Alex is going to take us through the subscription side of this, how it works, what you actually pay, and what typical usage is looking like a few weeks in. Over to you, Alex. Hi, my name is Alex Chisholm. I'm a product manager here at Posit, and I want to talk to you a little bit about the Posit AI subscription service. The Posit AI subscription service is a paid subscription that unlocks AI functionality within RStudio. So, it gives you access to what you've heard about relating to Posit Assistant and NextEditSuggestions.

There are a few different ways to get started. When you download a new version of the RStudio IDE, you'll see this new icon in the toolbar called Posit Assistant. When you click on that, you will see the new side panel open up, and you'll get a chance to look at Posit Assistant. You can sign in this way. It'll open up in your browser. You'll agree to some terms. You'll make a handshake, and all of a sudden, you'll be able to start working with Posit Assistant in RStudio as powered by Posit AI. The other way to get started is to go to Posit.ai, learn a little bit more about the service, and also start a free trial and create your account first, and then go back to the IDE and sign in to get started.

Either way, we want you to be able to maintain control over your data, and there are two pieces of information we want to share with you specifically related to this service. The first is, at the moment, Anthropic is the only large language model provider that we use to power Posit Assistant. We have a zero data retention agreement with Anthropic, which means your data does go to them for processing to give you a response, but then it'll be wiped from their system after that. And then on the Posit side, at the time of account creation, you'll get to set your data retention preferences. This defaults to not sharing. If you choose to opt in to help improve this service, we'll be able to look at your prompts and your trace data to make sure that the service is operating as expected, and maybe give us ideas for future iteration.

Understanding usage and costs

One important part of using Posit AI is understanding usage. Every time you use Posit Assistant, we're taking some prompt, some information of context from your session, we're sending it to Anthropic, they're going to perform some operations with the LLM, and it's going to come back with a response, or with code, or really whatever it is you requested. This counts as usage against your pro subscription, and each month you get a certain number of usage credits included within your plan. When that hits a limit, you can either choose to wait until the following month to continue to use Posit AI, or you can decide to opt in to what we call auto recharge. So on your billing page on Posit.ai, by default this is set to off, but if you do know that you want additional usage after you hit these monthly included limits, turning on auto recharge will enable you to constantly go back and charge you a bit more to continue your usage with the tools.

So if you were to turn this on, you're going to be able to then define exactly when this is going to add more funds to your account. This is completely reversible. If you decide to use this for a while, you can turn it off. You will also get an email every time that you do have an auto recharge, so that you have an understanding and awareness about what is being charged to your account.

Really the workload that you're doing on Posit Assistant is going to determine your usage levels, and this is very complex to get at because it is very case dependent. So it depends on the complexity of the task that you're working on, the amount of context and data maybe that you're providing to the LLM, the LLM model that you actually selected. So if you're talking Claude Haiku versus Opus, and also the length of the response that the LLM model is providing you back. All of these go into the usage that you have included within your account in a given month.

We have at this stage about a month worth of data on usage. We find the one thing is quite clear, if you do not use Posit Assistant and all you're using is next edit suggestions, these are provided in an unlimited way on the pro plan for Posit AI, so you would always be within this $20 monthly subscription. But also if you're using just lightweight Posit Assistant for occasional data work, cleaning things up, maybe you're not in it every day or only a couple days a week, we find that a lot of people are fitting within this $20 proscription level. If however you're using something daily, right, and you're doing it for your typical work, you know, we've seen people spend $20 to $100 including that base subscription plus those overages. And then real power users who are doing very, very large operations, maybe iterating through a bunch of data, putting in a lot of documentation, building packages, right. We see some people that are using this and they are coming up with bills that are, you know, $100 or more a month. And whether that is worth it for you completely depends on what you are doing with these tools and the purpose for them. Because what we are hearing from a lot of people is the level of productivity gains that people are able to get in RStudio are worth these expenses.

So that leaves me with like, who is Posit AI for? Again, Posit AI is just a subscription service that currently unlocks Posit Assistant and NextEdit suggestions in the RStudio IDE. In terms of who it is useful for, we're aimed at individuals right now that are comfortable with these data controls that we've put in place, both with that zero data retention agreement with Anthropic, but also the controls on data retention from Posit themselves. So if I think of who are some of these early adopters, we have people that are consultants or researchers or students and educators, people that generally are not working within an organization that has set up a fully controlled or sandbox environment for LLM work. I encourage you to take a look at Posit AI to learn more. And so you can understand and experience and see what so many people have provided feedback already on the benefit of using these tools. We understand that this setup might not be ideal for organizations or teams that need to have more control about where their data lives. And for that, we're going to learn more about Workbench and other opportunities to use these tools in more controlled environments.

Roadmap: Posit Workbench and enterprise features

Thanks, Alex. Now for the road ahead, and for the folks on the call asking about Posit Workbench, custom model providers, and what enterprise teams can do with all of this. Tom, take us home. Hi, my name is Tom Mock. I'm the product manager for Posit Workbench, Positron, and the RStudio IDE. I also oversee AI integrations into those tools, and I'll be talking a little bit about the next steps and roadmap for Posit Assistant inside of all three tools. I do always have to kind of provide an initial disclaimer. This kind of presentation portion here on roadmap contains four looking statements. So these are our goals and priorities, but the exact time frame is subject to change, as well as how specific features get implemented can change over time.

We have already had rich AI integrations into Positron. So both Positron Assistant and Databot will eventually evolve and be superseded completely by Posit Assistant as it moves into general availability. Posit Assistant can be an assistant or an AI agent and suggests and execute R and Python code directly in your session. So it combines some of the capabilities of both Databot and Positron that were previously separate experiences. It's ultimately built to automate the tedious parts of your development while keeping you, the expert, firmly at the wheel. Posit Assistant and our goal is to move into general availability in 2026 across RStudio, Positron, and inside of Posit Workbench.

As far as when I talk about Posit Assistant, I'll primarily be talking about it both in the context of Positron as well as RStudio. But if there are cases of it being unique to one IDE, I'll try to call that out for it. As far as differentiating real quick, we have Posit AI, the service. So this is a large language model provider that powers user interfaces and experiences across multiple tools. And Posit Assistant is a client software that provides the interface in both RStudio and Positron, as well as within Posit Workbench with those same IDEs. Posit Workbench provides enterprise features such as administrative controls or bring your own key, bring your own provider on top of Positron Pro and RStudio Pro and Posit Assistant within those IDEs.

As far as our first roadmap item, Posit Assistant will gain additional support for notebooks and notebook integration within Positron. On the right, we have the Positron Notebook Assistant that interfaces with our native notebook editor. And the notebook integration is targeted for integration with Posit Assistant in the 2026.06 release. Currently, you'd be making use of Positron Assistant with the notebooks until that release goes out the door.

This gives the assistant the ability to be aware of notebook session state, individual cells, as well as the ability to execute code in cells and see their outputs. It also helps with notebook editing in terms of it can make cell level changes or modify across the entire file to make improvements or modifications via AI agents. It also adds in conditional actions and next cell suggestions, as well as overall custom prompts. So you can say, do whatever you'd like with this notebook, as well as some of the predefined actions that we have seen users being asking for.

Moving on to Workbench and some of the enterprise roadmap here. In Q2, again, our goal of adding Posit Assistant on Workbench and adding enterprise features is initially bring your own key and bring your own provider support. All the enterprises that we work with are largely going to be expecting to bring their own existing large language model provider. We already have in general availability, Anthropic, AWS Bedrock, Snowflake Cortex, OpenAI and Posit AI, if that's available for you. And we also have preview support that just went out the door for Microsoft or what was formerly known as Azure Foundry, which also includes Azure OpenAI within that service and GitHub Copilot. Ultimately, our goal is to move providers that are within preview to general availability as well. And lastly, we have an experimental custom provider, which if you don't see your preferred vendor here on the list of GA or preview providers, custom provider allows you to bring a model provider that adheres to the OpenAI compatible endpoint or OpenAI compatible spec. So that opens it up to a lot of additional providers that at least have the completions and the slash models endpoint that allow us to power tools like Posit Assistant.

As far as moving into more of the Q2 and Q3 roadmap, again for features that Posit Assistant will gain on Workbench, we know that rolling out these tools requires specific kind of guardrails in some cases. So Workbench also has the ability to define global per user group and per user enforced administrative controls. Integrating these in the future with Posit Assistant will give us the ability to globally enable or disable Assistant or on a per user basis roll it out. Constrained to only approved LLM providers. Again, we have these goals and kind of existing support in some cases for enterprise providers that you already want to make use of and you want to avoid sharing your data accidentally or intentionally with other providers.

And additionally, part of our goal is to constrain to only approved customizations, where again, MCP servers or model context protocol servers can be very powerful for adding in customizations, but your enterprise may have an approved gateway or an approved registry, whether that's making use of Posit Connect to host MCP servers, or you have kind of a third party or external hosting service for those integrations. And then lastly, there are other extensions, just arbitrary extensions or AI tools that you might want to filter or exclude out. So with this integration, you'd be able to provide an inclusion list and say, these are the extensions I want to allow the users to install. And then users have kind of a set that they can install from, as well as you can host extensions inside your VPC or even on premise via Posit Package Manager. They recently added support for hosting a VSIX extensions that would work for both Positron and the VS Code IDE.

And lastly, on our enterprise features, we're looking at deeper auditing and utilization metrics. So again, as you roll these out, you might need to have awareness of things like who is doing these actions. Is it human executing code or is an agent executing code? Certain customers and regulatory industries need to measure those things for risk or compliance purposes. Or you might just want to understand kind of the ratio of human versus executed code over time as your team adopts tools that can execute code.

Additionally, for token utilization or uptake of Assistant, you might want to understand who's doing what. So who's making use of Assistant? How many tokens do I need to charge back to that user group? How many people are getting a lot of value from it? Can I go speak with them to see how Assistant is improving their day-to-day workflows? Just some of the core integrations around basic auditing and utilization metrics as we, again, look into the future for maybe deeper auditing capabilities per our user needs.

With that, I'll provide a brief summary. Again, Posit Assistant, our goal is to move to GA in 2026. Positron's Posit Assistant will gain deeper notebook integration in 2026.06. In Posit Workbench, we'll be adding broad enterprise provider support on top of both RStudio and Positron, administrative controls, and these auditing and other metrics that I just covered. Thanks, Tom. All right, let's open it up to your questions. Again, you can submit those through the Q&A panel in Zoom. Let's dive in.