Lotas Built a Coding Assistant for the 5 Million Data Scientists That Copilot Forgot

The Macro: The AI Coding Revolution Skipped an Entire Language

I have been covering AI coding assistants for over a year now and I keep noticing the same blind spot. Every major tool in the space, Cursor, Cline, Windsurf, Aider, and the rest, is built for software engineers writing JavaScript, Python, Go, Rust, or TypeScript. The models are trained heavily on those languages. The IDE integrations target VS Code. The workflows assume you are building software.

But there is an entire population of professional coders who do not build software. They build analyses. They write scripts that process experimental data, run statistical models, generate figures, and produce reports. They work in RStudio, not VS Code. They write R, not Python. And the AI coding assistant revolution has largely passed them by.

R has roughly 5 million active users worldwide, concentrated in academia, biostatistics, pharmaceutical research, epidemiology, and increasingly in quantitative finance and social science. These are serious professionals doing complex work. The R ecosystem, with packages like tidyverse, Bioconductor, ggplot2, and Shiny, is deeply specialized for statistical computing and data visualization. It is not interchangeable with Python for many research workflows. The people who use R use it for reasons, not because they have not heard of Python.

The coding assistant market is booming. Cursor has crossed $100 million in annual revenue. Copilot is embedded in millions of developer workflows. But when an R user opens RStudio and wants AI help writing a mixed-effects model or debugging a Bioconductor pipeline, they have almost no good options. They can copy code into ChatGPT, which works but breaks the flow. They can use Copilot in VS Code, but that means leaving RStudio, which is where their entire workflow lives. Neither option understands the R ecosystem deeply enough to be consistently useful.

This is a classic underserved market pattern. The mainstream tools do not care about this audience because it is smaller than the software engineering market. But 5 million users who are poorly served is a very large niche by startup standards. And data scientists tend to be loyal to tools that understand their workflow because so few tools do.

The Micro: An AI Agent That Lives Inside RStudio

Lotas, marketed under the product name Rao, is an AI coding agent built specifically for data science workflows in R. It lives natively inside RStudio. You do not switch editors. You do not copy and paste code into a web interface. The agent operates in the environment where R users already do all of their work.

Jorge Guerra is CEO and William Nickols is CTO. They are based in San Francisco, part of Y Combinator’s Summer 2025 batch, running a two-person team. The decision to build specifically for R and specifically for RStudio tells me they understand the user they are building for. An R user who has been asked to switch to VS Code knows exactly how much workflow disruption that involves. It is not trivial.

The product positioning as “the best coding agent for R” is aggressive and specific. It is not claiming to be a general-purpose coding assistant that also supports R. It is claiming to be the best tool for one language in one IDE. That kind of specificity either produces a product that deeply understands its user or a product that runs out of market. For a Summer 2025 YC company, I think the focus is right.

What I find compelling is the “agent” framing rather than “assistant.” An assistant suggests autocomplete and code suggestions. An agent suggests something that can execute multi-step data analysis workflows. The difference matters in data science because a typical analysis is not a single code block. It is a sequence of steps: load data, clean data, transform variables, run statistical tests, generate visualizations, format output. An agent that can execute that sequence while understanding the dependencies between steps is fundamentally more useful than one that suggests the next line of code.

The site mentions a “Lotas Pro” tier with a referral program, which suggests a freemium model. There is a download button for the product, indicating it runs locally as an RStudio extension or companion application rather than as a cloud service. Local execution is the right call for this audience because research data often has privacy constraints. Academic institutions and pharmaceutical companies are not going to send experimental data to a cloud API without extensive compliance review.

The trademark disclaimers noting non-affiliation with R Foundation and Posit Software are interesting. Posit, formerly RStudio the company, makes the RStudio IDE and is the most important player in the R ecosystem. Lotas building on top of RStudio without Posit’s endorsement is standard for third-party extensions, but it raises the question of whether Posit will eventually build competing functionality directly into RStudio. Posit has been cautious about AI integration so far, which gives Lotas a window.

The Verdict

Lotas found a genuine gap in the market. The AI coding assistant explosion has produced excellent tools for software engineers and almost nothing for data scientists who work in R. Five million users with an unmet need and high willingness to pay for tools that save them time is a strong foundation.

The competitive risk is twofold. First, Posit could build native AI features into RStudio. They have the distribution and the user trust. If they ship a good AI coding agent, Lotas becomes redundant overnight. Second, general-purpose tools could get better at R. Cursor and Copilot improve constantly. If their R support becomes good enough and R users migrate to VS Code, the RStudio-native advantage disappears.

In thirty days, I want to know how many active users are running the product daily. Not downloads. Active daily users who are writing R code with the agent assisting. That number tells you whether the product is sticky.

In sixty days, the question is whether the agent handles the hard parts of R. Easy R code, data loading, basic ggplot charts, simple t-tests, is not where users need help. They need help with complex statistical models, Bioconductor pipelines, Shiny app debugging, and the kind of obscure package interactions that produce cryptic error messages. If Lotas handles those well, it becomes indispensable. If it only handles the easy stuff, it is a novelty.

In ninety days, I want to see adoption numbers in academic labs and pharmaceutical companies. Those are the two largest R user bases and the environments where tool adoption tends to be sticky. A single lab that adopts Lotas and starts producing analyses faster is worth more than a thousand individual downloads because labs share tools and practices. One lab becomes five becomes a department. That is how developer tools grow in research environments.