What is structured scientific data?

Structured scientific data is data organized according to a defined schema, with consistent formats, labels, and relationships. In a lab context, this means a result has a defined type, a compound has a defined relationship to the assay it was tested in, and an experiment has a defined connection to the protocol it followed and the decision it was meant to inform. Data structured at the point of capture carries its context with it — the relationships that give a result meaning are recorded as a natural consequence of how the workflow is set up, not imposed afterward.

What is ontology-backed data in life sciences?

An ontology is a formal map of how concepts in a domain relate to each other. In life sciences, ontology-backed data defines not just what a piece of data is — a compound, an assay result, a material state, a protocol step — but what it means in relation to everything else in the system. This semantic richness travels with the information as it moves through the system, allowing AI to answer reasoning questions about scientific direction, not just retrieval questions about individual facts.

Why is it so difficult to retrofit unstructured lab data?

Retrofitting unstructured lab data faces three core obstacles. First, irreversibility: data never structured at the point of capture is often ambiguous in ways that cannot be resolved after the fact — the experimental context and intent are gone. Second, scale: retroactive structuring of years of scientific data competes directly with active research and commonly stalls before completion. Third, and most important, the underlying workflow that produced unstructured data is still in place, so new unstructured data continues to arrive even as old data is being cleaned. The data foundation is not a one-time project; it is an architectural decision about how scientific work is captured in the first place.

Why do AI initiatives in life sciences fail?

Most AI initiatives in life sciences fail not because of model limitations, but because the underlying data is not ready. Scientific data is often stored in free-text notebook entries, ad hoc spreadsheets, and proprietary instrument exports with no consistent structure or relationships. When AI is pointed at this fragmented foundation, it can produce plausible-sounding outputs but cannot generate verifiable, traceable, or trustworthy answers. Governance reviews stall because outputs cannot be audited. Gartner projects that 80% of agentic AI initiatives in healthcare and life sciences will not progress beyond initial governance checkpoints in 2026 for exactly this reason.

What does an AI-ready scientific data environment look like?

An AI-ready scientific data environment has five key characteristics: data is structured at the point of capture so scientists do not need extra steps to make it AI-ready; relationships between entities (compound to assay, assay to result, result to decision) are stored explicitly rather than inferred; experimental context travels with the data rather than sitting in notebooks or conversations; AI outputs are verifiable and traceable back to source data; and the platform is open enough for external tools and models to connect to the structured foundation without rebuilding underlying data bridges.

Introducing Luma Agent: Your AI Co-Scientist Is Here

Ben MorseMay 12, 2026

Luma Agent is now available. Scientists can describe what they need, and Luma will plan the work, execute it across their full scientific record, and return something without exports and without waiting for a specialist.

Luma has had AI capabilities for some time. Scientists have used them to connect to large language models, generate analytical summaries, develop visualizations, and get answers to questions within their dashboards. That foundation has been real, and it has served as a meaningful step forward for labs already working in Luma.

But the way scientists work is changing. Research questions are getting more complex. Datasets are getting larger. The overhead between a scientific question and a trustworthy answer is consuming more of the working day, not less. A reactive AI that waits for the next question has a ceiling. We have reached it.

Luma Agent: AI that acts, not just answers

Luma Agent is the next step in AI for Luma, and it is a significant one. It is not a smarter chatbot. It is a conversational AI that plans and executes complex scientific work on behalf of the scientist, using a suite of more than 30 specialized tools that understand Luma's data model, workflows, registration schemes, material states, and integrations.

The difference from what came before is not just depth. It is direction. Until now, Luma's AI was reactive: a scientist asked a question and got an answer. With Luma Agent, the AI is proactive. A scientist describes a goal. The agent builds a plan, works through it step by step, and comes back with something complete. The scientist describes the destination. The agent handles the journey.

What that looks like in practice: a scientist working in an Alzheimer's program runs R-group fragmentation analysis across 2,500 compounds in their application using a plain language prompt, receiving a structured report with key observations on R-group diversity and chemical space coverage at a 93.3% success rate, completed in minutes across 9 steps. A process that previously required a full workday of data exports and manual analysis, or 2 to 4 hours of specialist engineering time, now takes 1 to 5 minutes. A scientist preparing to close out an antibody discovery experiment asks the agent for a complete experimental writeup including full material lineage. The agent retrieves the task record, execution timeline, input and output materials, container configuration, and a full upstream dependency chain drawn from data already in Luma to return a complete, audit-ready report in 1 to 5 minutes. The same documentation effort previously took a full workday, and was frequently skipped entirely.

Luma Agent works because of what it operates on. Luma's data is structured at the point of scientific work, not cleaned up after the fact. Every write operation requires explicit human approval before it executes. Every step the agent takes is logged and auditable. In a market where Gartner predicts 80% of agentic AI initiatives in healthcare and life sciences will not progress beyond initial governance checkpoints in 2026, Luma Agent is built specifically to pass that gate.

What you need to know

Here is what is available now, who can access it, and how to get started.

For existing Luma customers

Luma Agent is available now. Connect with your Dotmatics account manager to enable the feature for your instance.
Once enabled for your account, you can access Luma Agent directly within the platform. No separate setup or data preparation is required. The agent operates on the structured data already in your environment.
If you’re configuring apps to run long-term tracking queries or monitor the changes to a research project over time, you can leverage Luma Agent to configure these Luma apps on your behalf. App configuration using the agent is approximately 50% faster than manual configuration.
Your account team can walk you through the three core use cases - complex data analysis, scientific exploration from the ELN, and automated experimental documentation - and help identify where to start based on your current workflows.

For scientists and digital lab leads evaluating Luma

Luma Agent is included as part of the Luma platform.
The fastest way to understand what Luma Agent can do with your data is to see it against a real dataset.
Key use cases to explore: self-serve correlation and fragmentation analysis, ELN-to-analysis workflows without manual data handling, and automated experimental writeups with full lineage.

For developers and data scientists

Gartner has explicitly recommended that CIOs require MCP-compliant interfaces from strategic R&D software vendors. Luma has this in production today. Most competitors do not.
Luma configures itself and its apps from the science you’ve already defined. When a new Luma app is needed, the agent doesn't wait for an administrator to translate scientific intent into schemas and SQL. It takes a plain-language description of the domain and builds the data model, columns, and relationships directly. The result isn't a blank form to fill in. It's a configured, ready-to-use application with auto-generated data flows and documented metadata, ready for the scientist to verify rather than build from scratch.

Built for where science is going

Luma Agent is not the end of the AI roadmap. It is the point at which AI in Luma becomes a co-scientist that acts on their behalf. The connective work that has always sat between scientific thinking and scientific insight can now be delegated. Scientists can focus on science. The agent handles everything in between.

We are committed to building the foundation that makes that shift trustworthy and durable: structured data, governed workflows, open protocols, and human approval at every decision point. That is what makes Luma Agent the right foundation as agentic AI matures.

Going to Bio IT World? Join us in the booth for a demo of Luma Agent.