XtalPi links molecular AI with the laboratory work needed to test it

XtalPi combines scientific software, drug discovery services and laboratory automation. Its attraction is the possibility of connecting molecular proposals to physical experiments in one research programme. That breadth also creates the main buying challenge: a buyer must identify which software, equipment, scientific work and development responsibilities are actually included in the proposed engagement.

01 / ProductA portfolio spans molecular reasoning and physical execution

The XtalPi discovery portfolio covers activities from hit discovery through lead optimisation and experimental support. Its computational approaches sit alongside screening, chemistry and structural work. The portfolio is relevant to AI because models inform what to make and test; the laboratory remains the mechanism for finding out whether those proposals behave as expected.

The digital chemistry page names several distinct tools. XMolGen addresses generative molecular design, XFEP addresses free-energy calculations and PatSight extracts chemical information from patents and publications. These functions should be evaluated separately. Extracting a structure from a patent is not an opinion on freedom to operate, and an affinity estimate is not a complete therapeutic profile.

XtalPi’s company account traces a progression from crystal-related computational work into broader AI discovery and robotics. That history helps explain the mix of offerings. It does not establish that every product can be accessed through one standard software licence or that every laboratory service is available in every geography. A meaningful enquiry starts by naming the desired output.

02 / AudienceDiscovery organisations and laboratory operators have different entry points

A pharmaceutical research group may seek a series of experimentally supported compounds against a target. Such a buyer needs chemistry and biology deliverables, a plan for negative results and rights to the resulting assets. It may value the connection between computation and experiments more than ownership of the equipment that performs them. The evaluation should follow the compound and its evidence through the programme.

A laboratory operator may instead want to automate an existing synthesis or screening workflow. It already has scientific objectives and needs reliable scheduling, sample tracking and instrument integration. For that buyer, a demonstration must address actual materials, reaction conditions and exception handling. A successful autonomous run on an unrelated chemistry set is a useful demonstration, but it cannot establish suitability for the intended process.

The Recursion blueprint offers context on integrating AI with experimental discovery at scale. The Schrödinger blueprint is useful for comparing a computational chemistry starting point. XtalPi’s breadth makes both comparisons relevant, but the selected engagement should determine which capabilities matter. Avoid comparing a whole-company narrative with a narrow software quotation.

03 / WorkflowProposed workflow: specify one design–make–test cycle in full

Choose one target or chemistry problem and define the experimental endpoint before selecting automation. This is a proposed evaluation workflow, not a Sequenced laboratory test. Write down the inputs, outputs and acceptance criteria for a complete cycle: available data, molecular proposals, synthesis, sample identity, assay measurements and the next design decision. The purpose is to expose gaps between the advertised components.

For molecular design, ask which constraints the selected tools can represent. Synthetic feasibility, selectivity and physicochemical properties may require different evidence. Preserve a record of why each molecule entered the experimental batch. Include comparators or controls that make results interpretable; a batch containing only novel high-scoring structures can leave uncertainty about whether a disappointing result reflects the model or the experimental setup.

The laboratory automation page describes XmartChem, ChemPlus and software for orchestration and scheduling. Confirm the exact instruments proposed for the work, their operating ranges and the handoffs requiring people. The same page marks NeoDispenser as coming soon. A named item in a broad product page should therefore not automatically be treated as available equipment in a committed delivery plan.

Run the proposed process through ordinary failures: unavailable reagents, an invalid measurement, an instrument interruption or a sample that cannot proceed. Decide who diagnoses each event and how the system records it. The benefit of automation depends partly on recovering trustworthy work after such interruptions. Throughput without traceability can simply create more results that the scientific team cannot confidently use.

Finally, review a completed cycle against a baseline the buyer already understands. Measure useful compounds or decisions delivered, scientist time required and the fraction of results needing rework. A laboratory can be highly automated while still optimising the wrong endpoint. Connect the operational evaluation to the scientific decision that justified the programme.

04 / PricingA public collaboration is evidence of a business model, not a tariff

XtalPi’s contact page provides routes for business enquiries. The reviewed pages do not supply a universal subscription price, workstation price or standard discovery fee. The product portfolio spans enough different deliverables that a single company-level price would be misleading. Request a proposal that identifies software access, equipment, integration, scientific services and support separately where applicable.

The Lilly collaboration announcement from June 2023 describes an agreement with up to $250 million in upfront and milestone payments. It assigns compound discovery work to XtalPi and subsequent clinical development and commercialisation to Lilly. The figure is potential deal value under that agreement, not a list price, cash already received or evidence that its undisclosed target has achieved a later clinical milestone.

For a new programme, define which costs are committed immediately and which depend on scientific or development events. For equipment, clarify installation, validation, consumables and maintenance. These are procurement questions, not published fee categories. No quotation was requested here, so an article cannot settle delivery dates, geographical availability or the terms available to a particular buyer.

05 / DistinctionsThe important distinction is continuity between models and experiments

XtalPi is notable because its documented offer crosses the boundary between computational discovery and physical laboratory execution. A buyer can investigate a workflow in which model output does not stop at a file of proposed molecules. The value of that continuity depends on whether information from the experiment returns in a form that can guide the next decision.

The breadth can also help reveal dependencies early. A change in molecular design may alter synthesis difficulty; a different assay may change what the model should optimise. Treat these as linked scientific choices. Buying each component independently can still be appropriate, but then the customer needs a clear integration owner and a reliable record connecting design assumptions with measured results.

06 / QuestionsAvailability and evidence need to be checked at product level

Which named modules are licensed software, delivered services or capabilities available only within a collaboration? What access exists to raw experimental data and the reasoning behind candidate selection? The public pages establish a broad portfolio but do not resolve these questions for each product. Ask for current documentation and a statement of work that matches the precise configuration being evaluated.

Drug discovery claims also need stage-specific interpretation. A candidate identified by a model, an active compound in a laboratory assay and a clinically effective medicine are different outcomes. Request the evidence for the particular stage being discussed, including unsuccessful experiments where useful. Do not use a collaboration’s maximum announced value as a proxy for scientific validation.

For automation, inspect sample tracking, instrument compatibility and how revisions to software or methods are qualified. Ask how a workflow can be transferred or supported if the original specialist leaves. The strongest demonstration is one the buyer can understand, repeat and maintain under its own operating constraints.

07 / DecisionSelect the smallest complete programme that answers a real question

XtalPi warrants consideration when a research organisation wants computational design connected to experimental execution, or when a laboratory needs to automate a well-defined scientific process. Start with a complete but bounded cycle. It should be large enough to expose the difficult handoffs and small enough that failure produces a useful decision without committing the entire research strategy.

Proceed when the scope identifies available products, named scientific responsibilities and measurable acceptance criteria. A broad vision of an autonomous laboratory is a starting point for discussion. A documented process that handles both successful experiments and ordinary failures is the basis for procurement and research planning.