Posit Assistant: Is it worth the switch?

What Is Posit Assistant?

Posit Assistant is an AI assistant built into your IDE, either Positron or RStudio. It’s built primarily for exploratory data analysis, and its pitch is session context and specialised, in-built skills.

If you follow Posit’s developments, you might have heard of their previous iterations of an integrated AI assistant. The naming has been a moving target:

• Positron Assistant came first, in early 2025, as a general coding assistant inside Positron
• Databot followed that August, aimed squarely at data exploration
• Posit Assistant replaces both, and is the first to be available in RStudio as well as Positron

In this blog, we will

• Take a look at a typical chat with Posit Assistant
• Explore what sets it apart from other integrated AI assistants
• Introduce the btw R package, for those who’d rather add R session context to their existing assistant instead

A Typical Chat

A great way to see it in action is to load some data and ask a broad question.

glimpse(happy)

happy |> 
  summarise(across(everything(), ~sum(is.na(.)))) |> 
  select(where(~. > 0))

drivers = select(happy, 
                `GDP per capita`:`Perceptions of corruption`)

cor(happy$Score, drivers) |> 
  round(2) |> 
  t()
#>                         [,1]
#> GDP per capita          0.79
#> Social support          0.78
#> Healthy life expectancy 0.78
#> Freedom                 0.57
#> Generosity              0.08
#> Corruption              0.39

Firstly, nothing in my request told it what the data represents, what columns existed or what the values looked like. Yet it came back with a description of the data, and a simple correlation analysis showing which variables drive happiness, and three suggestions of what you may want to explore next.

The other thing I like is that it answers by writing and then running code, which makes its decisions traceable. Language models are not designed to do maths (clue is in the “language” bit), but by writing code that does the maths, which we can scrutinise, copy and tweak, it gives us a much more trustworthy output.

How Does It Compare to Other Integrated Assistants?

Claude Code is probably the most popular alternative AI coding assistant, which is available inside Positron as an extension, and available anywhere you have a terminal through its CLI tool.

The two overlap on most of the basics. Both Posit Assistant and Claude Code:

• Can run on the same underlying model (though Posit Assistant also supports other providers)
• Can read and edit your files
• Generate and run code, from which they come to conclusions
• Can be given persistent project context files, through CLAUDE.md for Claude Code and AGENTS.md for Posit Assistant

So what sets them apart?

As we know now, Posit Assistant adds the session context. It knows the variables you have defined, the data frames you have loaded and the plot you are looking at, which is what made the exchange above possible.

Plots are one place where that makes a big difference. Posit Assistant can see the plots you previously made in the plot viewer, interpret them, and print plots back into the chat. That is worth a great deal in exploratory work, where much of what you learn comes from looking at a chart rather than a table. Claude Code can create a plot and save it to an image file, or ingest an image file, but the process is much clunkier.

It is not infallible, and Posit are candid about that: their research on how well LLMs interpret plots found models read straightforward charts well, but stumble when what is on screen contradicts what they expect to see.

It also brings the skills, which are bundles of instructions and code that the assistant pulls in by itself when it judges them relevant to what you asked. Thirteen ship built in, covering things like report authoring with Quarto and Jupyter Notebooks, and predictive modelling in R and Python, so it arrives at those tasks already knowing the conventions. You do not have to wait for it to spot the right one, though. Most skills are also exposed as slash commands, so typing /report, /quarto-authoring or /shiny-bslib runs that skill’s instructions there and then.

Claude Code is more autonomous, and will work through several steps before it checks in, whereas Posit Assistant stops for guidance more often. That is deliberate: it inherits Databot’s exploratory behaviour, so when it decides you are exploring data it runs a couple of tool calls, reports back and offers you some directions. The point is to keep you involved while the exploration is happening, rather than handing you a finished answer.

One thing to be aware of: Posit’s documentation lists “console history” as available context, which is a little misleading. Line by line detail of what has been run is not part of the context in either R or Python, so if you were hoping the assistant would know what code you’ve already run, or an error message that’s popped up, it will not.

Not ready to make the switch? Add context to your assistant with an MCP instead

If you would rather keep one assistant, it’s possible to give your AI assistant of choice some of the same session awareness with the btw R package, also developed by Posit. It works over the Model Context Protocol, and if MCP is new to you, Neal Richardson of Posit gave a good primer, “MCP, or Not MCP”, at our AI in Production conference.

The setup is straightforward, and the MCP setup guide walks you through it.

Once it is running, your assistant can inspect R objects and data frames in your session, check installed packages, and search for and describe CRAN packages.

There are trade-offs. It covers R only, so a Python session gets nothing from it. You also lose the skills, so your assistant comes to a Quarto document, Shiny app, or a modelling task without the guidance Posit Assistant would have loaded for you.

Closing Thoughts

Posit Assistant is easy to adopt, particularly if you already run the Claude Code extension in Positron or VS Code, because the workflow is familiar and the session context arrives without any configuration.

For exploratory data analysis it is an excellent tool. Being able to ask about the data you actually have loaded, and to read the code it wrote to answer you, changes the pace of that work more than I expected.

If you would rather keep to a single assistant, pair your go-to coding assistant with Posit’s btw package and you will get most of the way there.

This post grew out of “Improving Your Workflow with Positron and Claude”, a workshop we ran at AI in Production 2026. If sessions like that appeal, come and join us next year.

Find Out More

From us:

• A First Look at Positron and Posit Assistant - our free webinar on August 13th, 2026
• Positron vs RStudio - is it time to switch?
• “MCP, or Not MCP” - Neal Richardson (Posit) at AI in Production 2026

From Posit:

• A brief and biased history of Posit data science agents - Joe Cheng on why there were three of them
• Introducing AI in RStudio