From the Research Universities edition of September 26, 2026
Andromeda 2: evidence-grounded agentic lab hits 50% high-AUC formulations vs 17% / 2% baselines (paclitaxel)
Intrepid Labs (Toronto) — arXiv preprint · Industry autonomous laboratory / pharmaceutical formulation R&D · Canada (Toronto); transferable methods with limits
- Publisher
- arXiv 2609.19099
- Original publication
- September 16, 2026 (arXiv)
- Source retrieved
- 2026-09-27
What happened
Andromeda 2 is an agentic formulation system that reasons over structured in-house experimental evidence and drives a miniaturized autonomous SEDDS lab. At a matched budget of 96 paclitaxel formulations, it achieved a 50% high-performance hit rate versus 17% for probabilistic Andromeda 1 and 2% for wet-lab DoE; 12 vs 6 vs 0 formulations met all four TPP objectives. Median AUC10–240 was 70.1 / 12.0 / 3.5 mg·min/mL. Ablation: withholding historical in-house evidence cut mean AUC by 34% (62.0→46.2). Authors note FaSSIF is not bioavailability; single campaign limits inference; industry lab setting.
Why it matters
University and national-lab AI-for-science programs evaluating autonomous labs need measured hit-rate gains from evidence grounding—not agent demos alone—and must separate in-vitro assay gains from clinical claims.
Evidence and measured results
arXiv abs 2609.19099 opened via HTTPS; hit rates, TPP counts, median AUCs, and 34% ablation taken from abstract/abs page text (paper dated 16 Sep 2026).
Limitations and uncertainty
Industry (Intrepid Labs) not a university lab; FaSSIF ≠ oral bioavailability; single campaign per arm; preprint not peer-reviewed journal version; paclitaxel-specific design space.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-09-26; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Formulation and autonomous-lab teams spend scarce wet-lab budget without learning from prior experimental evidence.
- Stakeholders
- VPR / research computing for AI-lab platforms
- Core facility directors
- PI formulation / chem teams
- Research integrity / data governance
- Discovery questions
- Is historical experimental evidence structured and reusable across campaigns?
- What matched-budget baselines exist (DoE, Bayesian optimizer, prior agent)?
- Which TPP metrics are assay-valid vs translational?
- Are models and evidence kept in-house or via external APIs?
- Value hypothesis
- Evidence-grounded agentic design can raise high-performing hit rates at fixed wet-lab budget versus probabilistic optimization or DoE when historical evidence is available.
- Potential engagement
- Matched-budget pilot on one hard API with ablation of historical evidence and pre-registered TPP.
- Unsupported claims
- In vivo bioavailability improvement
- Guaranteed gains on every API
- Peer-reviewed clinical endpoints
- University-lab identical ops cost
Pre-sales engineering
Role takeaway
- Fit
- Agentic proposal/critique loop over constrained SEDDS design space with robotic execution and feasibility engine.
- Architecture and integration
- Wire evidence DB, assay LIMS, composition validator, and batch review UI; keep labels/metrics outside unconstrained agent writes where possible.
- Prerequisites
- Structured historical experiments
- Robotic liquid handling
- Defined TPP assay
- Feasibility constraint engine
- Human batch review SOP
- Constraints
- Industry preprint setting
- Single-campaign statistics
- FaSSIF endpoint only
- API-specific design space
- Security
- Protect formulation IP and experimental databases; decide on-prem vs external LLM APIs before campus deployment.
- Proof of value
- Matched-budget hit rate vs baselines
- Full-TPP count
- Evidence ablation delta
- Replicate campaign if claiming generality
Delivery
Role takeaway
- Work
- Implement evidence schema, agent tools, feasibility checks, and evaluation harness; run matched-budget pilot.
- Dependencies
- Lab robotics
- Assay analytics
- Data engineering for evidence store
- Scientific reviewers
- Ownership
- Platform team owns robots/software; PIs own scientific claims; research integrity owns disclosure of AI role.
- Skills and adoption
- Train scientists to set TPPs and audit agent rationales; train operators on batch review without silent redesign.
- Governance checkpoints
- Pre-register TPP and budget
- Ablation plan review
- Endpoint interpretation review (no PK overclaim)
- Acceptance criteria
- Documented hit-rate vs baseline at fixed N
- Evidence ablation completed
- Human review logs retained
- Risks
- Assay misread as clinical success
- Evidence store biased to past APIs
- Single-run variance
- Vendor lock on closed evidence
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Agentic planner + deterministic feasibility engine + robotic wet-lab loop; historical evidence store separate from campaign measurements; human scientist reviews batches without redesigning compositions.
Governance
Who approves, reviews and stays accountable for outcomes?
Evidence provenance and ablation transparency matter for claiming agent value; university transfer needs data-governance for reusable experimental evidence across PIs.
Security and privacy
What data, permissions and controls need testing?
Formulation IP and experimental databases are sensitive; paper uses fully in-house models (no external commercial API), which is a deployment choice campuses may or may not mirror.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Raises need for data-curation and assay specialists who can maintain reusable experimental evidence—not only GPU engineers.
Procurement
What should contracts, pricing and exit terms secure?
Buy or build evidence platforms and evaluation harnesses with ablation requirements; do not procure agent seats without matched-budget acceptance tests.
Operating model
Which teams own the service once it runs?
Shift from one-off campaigns to accumulating institutional experimental evidence that improves subsequent agentic search—requires cross-PI data sharing norms.
Publication history
- 2026-09-26Research Universities · Issue 204 resources
Stable resource ID: andromeda2-evidence-grounded-autonomous-lab-20260916