AI-native contract research organizations: deliverables compared

An AI-native contract research organization can sell a synthesized compound, a bioanalytical result or clinical trial execution. A software license and a therapeutic asset license buy different rights and responsibilities. Warp, Dash Bio, Harbor and Lindus illustrate those differences, even when their descriptions share the same AI vocabulary.
Compare the work a buyer accepts and pays for. Warp publishes a per-delivered-compound model. Dash publishes sample-analysis prices and retesting terms. Harbor's current site sells electronic data capture (EDC) software, while its founder previously announced a full-service CRO contract. Lindus now describes a biopharma business that in-licenses and develops therapeutic assets.
Sources were checked on 16 November 2026. The comparison records company statements, not verified service performance. It is a selected public-source comparison, not a ranking or a census of AI-enabled CROs.
Delivered work and current scope
| Offering | Delivered unit or development aim | Public scope | Evidence limit |
|---|---|---|---|
| Warp chemistry service | Delivered compound | Preclinical small-molecule lead-optimization analog synthesis, mg to multi-gram scale | Small design-partner group; published terms do not establish comparative reliability |
| Dash Bio bioanalysis | Sample results, developed method or validated method | ELISA, MSD, Luminex, LC-MS, qPCR and ddPCR | Published service specifications and prices, not independent trial-wide outcomes |
| Harbor EDC | Software license and study setup support | Draft study builds, source extraction and monitoring assistance | Current software offer; human review remains part of the workflow |
| Harbor's announced trial execution | End-to-end trial execution contract | Founder announced a medical-device trial contract in May 2026 | Current full-service availability is not confirmed by the reviewed EDC pages |
| Lindus clinical execution case | Trial execution through clinical study report | Company reports a completed Phase 2a ME/CFS study | Historical execution evidence; not a current service quote |
| Lindus asset development | Advancing in-licensed therapeutic assets | Integrated clinical development toward value inflection points or approval | Company strategy, not a disclosed licensing agreement |
The same company can appear twice because a past service engagement and a current business model answer different procurement questions. Harbor's founder announcement supports the claim that the company signed a trial-execution contract. It does not establish successful delivery or prove that every current EDC buyer can purchase that service. Lindus's clinical case documents its account of work performed. Its current homepage leads with asset development.
Acceptance, payment and reruns
| Offering | Acceptance disclosed publicly | Billing and rework disclosed publicly | Still needed in a contract |
|---|---|---|---|
| Warp | Compound delivery; numeric purity, identity and quantity thresholds not disclosed in reviewed pages | Customer-agreed price per delivered compound; if Warp does not produce it, customer does not pay | Release specification, failed-route treatment, delivery deadline and rerun terms |
| Dash Bio | Method development covers assay characteristics; project-specific numeric limits not disclosed | Published sample prices include required retesting; validation guaranteed or no charge; batch minimums and extras apply | Approved method, acceptance limits, repeat rules and full project quote |
| Harbor EDC | Team reviews and tests study build and confirms extracted values | Commercial study licenses start at $2,000/month; enterprise volume pricing | Study UAT sign-off, correction duties, export format and termination plan |
| Harbor announced trial execution | Contractual acceptance criteria not disclosed | Founder announced a seven-figure contract, without a billing breakdown | Statement of work, pass-throughs, payment triggers and failed-milestone terms |
| Lindus clinical case | Reports protocol and regulatory/ethics approvals; payment acceptance not disclosed | Reported study cost is not a public rate card; rerun terms not disclosed | Service availability, current quote and agreed deliverables |
| Lindus asset development | Deal-specific milestones not disclosed | Fees, royalties and failed-study allocation not disclosed | License scope, development duties, milestones and termination rights |
Sources: Warp service model, Dash pricing, Harbor company page and pricing, Harbor founder launch, Lindus case and homepage. Prices are the vendors' public statements on the check date, not Arca quotes.
Dash's page makes the scope of an advertised price worth checking. It lists assay-dependent batch minimums, a GLP/GCP surcharge, tissue-homogenization fees and charges for additional analytes. Its qPCR and ddPCR listings specify a 20-sample minimum, but its FAQ says at least 30 samples for any sample-analysis project. A buyer needs the applicable minimum confirmed in the quote. Required retesting being included does not make every project configuration the same price.
Warp states that its typical synthesis work requires 5 to 15 synthetic steps. Paying per delivered compound transfers some risk of unsuccessful production to the supplier. The reviewed pages do not disclose the purity threshold, analytical package or deadline that makes a particular delivery acceptable. Those remain contract questions.
Data rights and therapeutic ownership
| Offering | Public data handoff or reuse statement | Public program or asset position |
|---|---|---|
| Warp | Describes learning from experiments and running dedicated experimental campaigns; customer data rights and training permissions not disclosed in reviewed pages | Says it is not developing drugs of its own; customer IP assignment clauses not disclosed |
| Dash Bio | Reporting and data transfer included; ownership and training/reuse permissions not disclosed in reviewed pricing page | Asset ownership and IP assignment terms not disclosed in reviewed pricing page |
| Harbor EDC | Standard exports included; ownership and training/reuse permissions not disclosed in reviewed offering pages | Asset ownership and IP assignment terms not disclosed in reviewed offering pages |
| Harbor announced trial execution | Data ownership and training/reuse permissions not disclosed in reviewed launch | Asset ownership and IP assignment terms not disclosed in reviewed launch |
| Lindus clinical case | Customer data rights and training/reuse permissions not disclosed in reviewed case | Asset licensing terms not disclosed in reviewed case |
| Lindus asset development | Partner data rights and training/reuse permissions not disclosed in reviewed homepage | In-licenses and develops assets; exclusivity and retained rights not disclosed |
Sources: Warp's stated strategy, Dash pricing, Harbor pricing and founder launch, Lindus case and current positioning.
“Not disclosed” means the reviewed source pages do not settle the field. It does not mean a company lacks contract terms, takes customer IP or trains on customer data. Website privacy notices are not substitutes for negotiated research-data rights.
Warp's ambition to learn from experimental evidence does not establish permission to reuse a customer's confidential results. Lindus's in-licensing strategy does not tell a potential partner which territory, indication or development rights a particular deal would transfer. Ask for the clauses rather than inferring them from a company label.
Worked example: accepting an assay result
Consider an illustrative procurement exercise, not a Dash or other vendor protocol. A sponsor needs a concentration dataset for 30 plasma samples using an established single-analyte method. The supplier has completed every instrument run and returned a spreadsheet. Procurement still needs to establish whether the sponsor can accept the work.
| Proposed purchase requirement | Evidence the sponsor would request | Contract decision |
|---|---|---|
| Analyze the intended samples with the approved method version | Sample manifest, method identifier and run records | Define who approves sample substitutions or a method change |
| Return results that meet the agreed assay rules | Calibration and QC outcomes, dilution history and flagged results | Agree numeric acceptance limits with qualified assay staff before work starts |
| Resolve failed or unreportable measurements | Documented reason for each repeat, missing result or out-of-range value | Define included repeats, extra charges and when a fresh sample is required |
| Deliver a usable dataset | Units, sample identifiers, analysis version and report with exceptions | Specify file formats, reconciliation checks and reviewer sign-off |
| Preserve the sponsor's intended rights | Executed data/IP and confidentiality clauses | Set reuse permission, retention, access and exit obligations |
No clinical or assay acceptance thresholds are proposed here. Qualified staff must set them for the method and intended use. A completed run can yield an unreportable measurement. An acceptable assay dataset can still fail to answer the program's biological question. A service contract should identify which of those outcomes triggers payment and what happens next.
Dash's public inclusion of required retesting answers part of the commercial question. It does not replace a study-specific method or acceptance agreement. The same distinction applies to compound delivery and trial execution: specify the deliverable, its review evidence and the unresolved scientific question separately.
Faster operations and scientific evidence
Dash CEO Dave Johnson's March 2026 essays describe queues and handoffs and outcome-based billing. They explain the company's operating and commercial rationale. Their models are not measurements of every CRO or proof that changing a billing unit removes waiting time. A buyer can test the claim by asking for the interval between sample receipt, analysis, quality review and accepted data delivery, including repeat work.
Lindus's 4 November case reports a trial cost just below $1 million. It describes 25 randomized participants, a crossover design, nine weeks of participation and a single virtual site. Its comparison uses modeled GlobalData budgets rather than observed matched trials. The account supports a specific company-reported execution example. It does not isolate AI's contribution or establish the cost of a different trial. The clinical trial cost breakdown separates that reported budget from assumptions for another study.
Faster source extraction, a faster synthesis or a cheaper completed trial does not itself establish that a medicine works. The purchased operational output and the evidence needed for the next scientific decision should each have a named reviewer.
Written clinical responsibilities
CRO outsourcing is not legally mandatory for a US investigational drug trial. 21 CFR 312.52 says a sponsor may transfer any or all Part 312 obligations to a CRO. The transfer must be in writing. Obligations outside that written description remain untransferred. A CRO that assumes an obligation must follow the applicable regulations and faces the same regulatory action as a sponsor for failure to comply with it.
That provision concerns investigational drugs under Part 312. It is not a blanket transfer rule for every device study, chemistry service or software subscription. Purchasing an EDC license does not by itself document a transfer of sponsor obligations. For a clinical engagement, map the actual written duties and retained sponsor work. “Full service” and “AI-native” do not identify those duties.
Download and method
Download the source-dated deliverables matrix (CSV). Each row records one offering and one field, its disclosure status, source URL, any displayed publication date and the date checked. Separate rows preserve Harbor's announced service and current software offer, and Lindus's execution case and asset strategy. Blank publication dates mean the source does not display one.
The matrix is a manual extraction of selected first-party pages. No operators were interviewed, no private contracts were reviewed and no services were tested. Unknown fields are bounded to the pages cited. The reproducible export script generates the CSV and checks it for drift. It does not independently verify vendor statements.
For regulatory and quality software rather than outsourced experimental work, see IQVIA alternatives. For the document review work around clinical delivery, see the clinical study report glossary.