An infrastructure that allows organizations to implement AI on their own terms

[#1]

There is no shortage of noise in the AI conversation right now. Every week brings new terms and acronyms – agentic AI, reasoning LLMs, vibe coding – and with them, a new wave of pressure on organizations to declare an AI strategy or adopt a platform, all under the implied risk of being left behind. Scientists and engineers using JMP are not immune to this pressure. They are hearing it from their leadership, and they are asking us what it means for the tools they rely on every day.

Our answer is straightforward: JMP's core value has not changed, and it will not change. What has changed is our ability to make that value more accessible, more discoverable, and more extensible – and our AI infrastructure is a meaningful part of how we do that. But we are doing it deliberately, on your terms, not ours.

Why JMP's analytic workflow remains irreplaceable

Before discussing where AI fits, it is worth being precise about what JMP does that no language model can replicate.

JMP was designed from the ground up to be interactive. A user drags a column into Graph Builder and sees their data instantly – no code, no syntax, no prompt. Click a point on a graph and it’s highlighted in the data table and every other graph across the entire analysis session. But this interconnectedness not a feature; it’s a philosophy. Simply stated, the scientist or engineer, armed with domain expertise and the right analytical environment, is the engine of discovery.

The JMP analytic workflow moves left to right. It starts with a person and a problem. JMP translates business questions into analytical ones, guides users to the right methods, and provides an interactive, visual environment where insight happens quickly. JMP Live then extends that value across the organization – making work visible, sharing proven methods, and moving analysis from individual laptops to enterprisewide collaboration.

Large language models do not sit at the center of this workflow. They plug in at specific points, and they make parts of it faster. The scientist or engineer still drives the analysis, still provides the value.

The curve that explains everything

There is a useful way to visualize why JMP and LLMs are not competitors. Imagine a chart with problem complexity on the X axis, with common, well-documented problems on the left and rare, high-value problems on the right.

LLM training data volume drops off sharply as problems become more complex and rarer. LLM effectiveness follows. Common problems, such as how to write a basic Python script or how to debug a common error, have thousands of examples online, and LLMs excel at them. But as you move toward the right side of the curve, toward novel experimental designs, complex multivariate analysis, and problems that no one in your industry has solved before, LLM effectiveness collapses. There simply is no training data for results that do not yet exist.

JMP's effectiveness moves in the opposite direction. JMP gets more valuable as problems get harder and rarer. Design of experiments, exploratory data analysis, discovery at the frontier of what is not yet known – this is JMP territory. The thesis statement is worth stating plainly: training data abundance does not equal business value. LLMs are trained on what is common. The problems worth solving for your organization are the ones with the least training data available.

You cannot prompt an LLM to, say, "Design the next blockbuster drug and make no mistakes." JMP users work on the edge of what is not yet known, and that workflow cannot be automated away with AI alone.

Where LLMs genuinely add value in the JMP ecosystem

With that foundation established, there are real and meaningful opportunities where LLMs accelerate the JMP workflow.

Decreasing time to value for new users. JMP has a short but steep learning curve. A beginning user who knows their science but not yet the software can hit a wall in the first few hours. LearnBot (available on the JMP Marketplace) addresses this frustration directly. By loading JMP's rich learning content, documentation, and the JSL Scripting Index into an LLM-augmented knowledge base, LearnBot lets new users ask their questions in natural language and receive a response grounded in accurate, up-to-date JMP content. Early results have been strong.

Closing the gap between point-and-click use and automation. JMP's scripting language, JSL, is extraordinarily powerful. It can build custom applications, integrations, and organization-specific tools. But telemetry data tells us that the vast majority of JMP users never write it; even the most basic JSL activity, saving a script to a data table, is used only 16% of the time. That underutilization is a significant bottleneck. A user can describe what they want in plain language and receive working JSL in return; the automation and customization that JSL enables becomes accessible to the entire user base, not just the small group of experts that an organization typically relies on.

Making JMP's hidden capabilities discoverable. Organizations around the world have relied on JMP’s value for more than three decades. But sometimes, JMP’s capabilities can be hidden behind red triangle menus or platform options that most users never find. Natural language interfaces make these tools available to everyone. A user can describe what they want to do and be pointed to the right tool within JMP, even if they never would have found it in the menu.

Enabling freeform text analysis within the JMP workflow. The LLM Prompt Runner extension allows users to pass data table content directly to an LLM and return structured results (such as sentiment analysis, root cause extraction, or classification) back into the data table. Users stay in the JMP workflow while leveraging LLM capabilities for the tasks where they genuinely excel.

The infrastructure we are building

JMP is being built for maximum flexibility, not betting on one vendor. We are not assuming a single LLM provider will win. Our focus is on infrastructure and interfaces that allow customers to connect JMP to the models and services that align with their standards and environment.

Here is how that comes together:

First, the knowledge base. We have published more than 20,000 JSL script snippets at JSL.jmp.com in formats that are easy for LLMs to interpret, with 10,000 more on the way. As a result, LLMs are getting significantly better at understanding and writing JSL, and they will continue to improve.

Second, the translator layer. We use retrieval-augmented generation (RAG) to minimize hallucinations and ensure that when a user asks a question, the response is grounded in accurate, relevant, and up-to-date JMP content.

Third, the plumbing. We are making it easier to connect JMP with the LLM of your choice through endpoint connectors, so customers are not locked into any single provider. It’s coming in March 2027, and if you are interested in trying it out first, please join the JMP Early Adopter Program; we would love your feedback.

Fourth, tools. JMP continues to build a library of high-value AI and ML extensions that support users, improve their workflows, and reduce the need to code. Users can install these extensions, configure them to align with their organizations’ policies, and modify them as needed.

And finally, the JMP Marketplace. Through the JMP Marketplace, customers choose which LLM or AI extensions make sense for their needs. By offering these extensions through the JMP Marketplace rather than embedding them in the core product, JMP keeps AI optional, not forced.

Security, privacy, and compliance – on your terms

We understand that data privacy, security, and compliance are top of mind, especially when AI is involved. JMP has made it straightforward to find the answers your security and compliance teams need.

Our Trust Center covers what JMP does with your data and prompts, how we ensure data privacy, what security assurances we provide, and how we address EU compliance requirements. The JMP Generative AI Terms detail specifically what happens with your prompts when using LLM-powered extensions. Each extension also includes a Model Info Sheet that identifies which model is being used and links to the model provider's documentation. And our JMP Responsible Use Policy outlines our commitment to ethical AI practices and guides users through both permitted and unacceptable use cases.

All of these resources are public and available for you to share with your security and compliance teams.

What JMP is not doing

It’s just as important to clarify what JMP is not doing as it is to explain what we are doing.

• We are not building a chatbot to replace the JMP analytic workflow. JMP is an analytics tool, not a conversation product.
• We are not letting LLMs do the math: JMP computes, LLMs write the instructions.
•  We are not training or fine-tuning foundation models since that is a billion-dollar GPU game, not a desktop analytics play.
• We are not generating synthetic data. JMP's brand is statistical rigor, and that rigor will not be compromised by LLM-fabricated numbers.
• We are not forcing AI features on users who do not want them.

The strategy is the business outcome and the workflow. AI is one tool in the toolbox to get there.

The bottom line

JMP has a strong foundation and a differentiated workflow. The analytic workflow (visual exploration, statistically sound methods, repeatable and shareable analysis) is not the same thing as typing a prompt into a chat box. A prompt can assist, sometimes dramatically. But it does not replace the rigor and the end-to-end workflow that JMP enables.

What we are building is infrastructure: a knowledge base that makes LLMs significantly better at understanding JMP, a translation layer that grounds responses in accurate content, endpoint connectors that give customers flexibility in choosing their AI providers, and a marketplace that delivers AI capabilities optionally and on your terms.

JMP users create new knowledge. They work on problems that have never been solved, with data that no model has seen. That workflow is not going away. If anything, in an era of AI, it becomes more important, because the problems worth solving are precisely the ones where LLMs run out of training data and JMP's strength begins.

We are not chasing hype. We are building for the long term, and we are building for you.