Research that earns its citations.
Praxen Institute publishes governance, evaluation, and policy work in the disciplines that the rest of our practice depends on. The standard is peer-reviewable, the audience is researchers and policy readers, and the format is whatever the argument requires — a paper, a benchmark, a dataset, an essay.
Active programs
Agentic Real Options. A framework that prices AI technical debt as the option-cost of deferred capability-acquisition — borrowing the valuation machinery of financial real options to make concrete what most teams treat as gut-feel. Worked example: migrating a retrieval pipeline from lexical BM25 to a learned dense retriever is priced as the exercise of an option against the latency-budget strike, with the deferral decision rendered a function of corpus drift rate and the volatility of downstream eval metrics. arXiv-track working paper, affiliation: Praxen Institute. Status: revision in progress.
Self-Driving Laboratories. Closed-loop systems in which generative models, agentic coordinators, and automated wet- or dry-lab instrumentation together run one discovery loop — Coscientist (Boiko et al., Nature 2023), ChemCrow (Bran et al., Nature Machine Intelligence 2024), RFdiffusion-designed antibodies confirmed atomically by cryo-EM (Bennett et al., Nature 2025), SyntheMol's six novel antibiotics against A. baumannii — are each pieces of what is becoming the standard stack. The program's angle is governance: how do you peer-review work whose researcher is an agent swarm with gpu-months of planning credits, and where does the option-value framing from Program 1 price the tradeoff between a narrower targeted wet-lab run and a broader in-silico screen? The 2024 challenge to the Ceder-group A-Lab synthesis claims is the canonical case study. Status: field study active.
Publications
Contact
Research correspondence and collaboration inquiries: institute@cld-dev.io.