{"slug": "thirstday-7-where-can-i-order-the-experiment-my-ai-just-proposed", "title": "THIRSTDAY #7: WHERE CAN I ORDER THE EXPERIMENT MY AI JUST PROPOSED?", "summary": "Anthropic reported on September 23 that roughly 950 Claude agents worked for 21 hours across sequence data and helped researchers identify an unusual enzyme system, after which human scientists continued with analysis and laboratory experiments. Anthropic cautions that the biological function is still being studied, so the result is not evidence that AI discovered the next CRISPR; the company says it shows large amounts of digital scientific reasoning can be compressed into a short period before laboratory validation. The episode highlights the remaining gap between a testable scientific idea and a laboratory-ready job, a translation step that services such as Emerald Cloud Lab and Science Exchange already partially address.", "body_md": "Anthropic published an unusually good example of that collision on September 23. According to the company, roughly **950 Claude agents worked for 21 hours** across sequence data and helped researchers identify an unusual enzyme system, after which human scientists continued with analysis and laboratory experiments. Anthropic is careful about the conclusion because the biological function is still being studied, so this is not “AI discovered the next CRISPR”; what it does show is that large amounts of digital scientific reasoning can now be compressed into a relatively short period before somebody still has to enter a laboratory and test whether the interesting idea survives contact with reality.\n\nFor a large research organization, that transition is familiar because the laboratory, procurement staff, protocols and specialist knowledge may already exist somewhere inside the institution. For a two-person startup, an independent researcher or a technically capable community, however, the same moment raises a more practical question: **if my AI has produced a scientifically credible proposal, where do I take it next?**\n\nImagine that an AI has read the relevant literature, inspected public datasets, compared several mechanisms and arrived at something like this:\n\n***“Run this assay under these conditions, compare these variants, and the result should distinguish hypothesis A from hypothesis B.”***\n\nThat sounds close to a result when it appears inside a chat window, although in physical science it is closer to a procurement request.\n\nSomeone still has to decide whether the proposed experiment is valid, which controls are necessary, what samples and reagents are required, whether the chosen equipment can measure the effect, how much material is needed, which safety rules apply and what would count as a usable outcome. Even a routine experiment contains details that a plausible paragraph can glide over but a laboratory cannot.\n\nThe interesting gap therefore sits between **a testable scientific idea** and **a laboratory-ready job**.\n\nWe should also avoid pretending that nobody has tried to close it, because remote laboratories, CROs and research marketplaces already exist.\n\nEmerald Cloud Lab offers remote access to automated and manual laboratory capabilities, including one-off experiments as well as longer research programs. Science Exchange takes a broader marketplace approach, connecting research organizations with qualified suppliers while handling sourcing, contracting, compliance and payment across a large network of scientific services.\n\nSo “why can’t I order an experiment online?” is already the wrong question.\n\nA better one is **how close we are to making the whole journey predictable enough that a small team can use it without first becoming experts in laboratory procurement.**\n\nSoftware has trained us to expect a fairly clean interface between intention and execution. We describe the workload, choose resources, receive a price, run it and inspect the output.\n\nPhysical research contains more translation.\n\nA model might propose “compare the activity of these three variants,” while the provider needs to know the expression system, purification requirements, assay conditions, controls, plate layout, detection method, sample quantities and acceptable error. If any of those decisions change the scientific meaning of the experiment, then the translation step is not administrative overhead; it is part of the research itself.\n\nThat is why the useful product is probably not a button labelled **RUN EXPERIMENT**. A better system would take the AI-generated proposal, expose its assumptions, identify missing protocol details, route it through scientific and safety review, find facilities that can actually perform the work, and return quotes that are comparable enough for a small team to make a decision.\n\nThe desired workflow looks much more like this:\n\n```\n┌──────────────────────────────────────────────────────────────┐│                     RESEARCH QUESTION                        ││  What observation would distinguish the competing ideas?    │└──────────────────────────────┬───────────────────────────────┘                               │                               ▼┌──────────────────────────────────────────────────────────────┐│                     AI RESEARCH PROPOSAL                     ││  Literature, data, suggested experiment, assumptions,       ││  controls and expected observations                         │└──────────────────────────────┬───────────────────────────────┘                               │                               ▼┌──────────────────────────────────────────────────────────────┐│                     FEASIBILITY REVIEW                       ││  Is it scientifically valid, safe and executable?           ││  Which protocol details or materials are still missing?     │└──────────────────────────────┬───────────────────────────────┘                               │                               ▼┌──────────────────────────────────────────────────────────────┐│                  PROVIDERS · PRICE · SCHEDULE                ││  Suitable lab or CRO, total quote, turnaround time,         ││  required materials and preparation                         │└──────────────────────────────┬───────────────────────────────┘                               │                               ▼┌──────────────────────────────────────────────────────────────┐│                         EXPERIMENT                           ││  Protocol version, execution record, controls, deviations   ││  and instrument outputs                                     │└──────────────────────────────┬───────────────────────────────┘                               │                               ▼┌──────────────────────────────────────────────────────────────┐│                     RESULTS + EVIDENCE                       ││  Raw data, processed outputs, provenance and enough detail  ││  for another qualified lab to understand or reproduce it    │└──────────────────────────────────────────────────────────────┘\n```\n\nThe diagram looks almost boring, which is probably a good sign, because the missing convenience is not magical autonomous science; it is **a reliable interface between reasoning and physical execution**.\n\nEmerald Cloud Lab is interesting because its model resembles cloud infrastructure more closely than traditional contract research. Experiments can be specified remotely and executed inside a standardized facility, while the platform advertises access ranging from individual experiments to larger programs.\n\nScience Exchange solves another part of the problem by making the provider landscape easier to navigate. Its network covers discovery work, preclinical services, biological materials and specialist laboratory capabilities, while procurement and contracting are handled through a common commercial layer. The company says more than **$1 billion in historical R&D transactions** has moved through its platform, which at least tells us that outsourced scientific work is not a theoretical market.\n\nHowever, the existence of those services does not mean that an AI-native startup can submit an arbitrary scientific proposal tonight and receive three transparent quotes tomorrow morning.\n\nDifferent providers support different techniques, some work depends on geography or controlled materials, protocols may require substantial scientific refinement, and many services are still quote-driven rather than priced like commodity compute. Waiting time can also matter as much as price: a $2,000 experiment available next week is a different product from a $1,200 experiment available in three months.\n\nThis is where I think the next useful layer may emerge, because **AI can reduce the cost of reaching a good question faster than the current market reduces the cost of arranging the physical test**.\n\n**AI-assisted literature and data exploration.** Models can already help researchers search, compare, summarize and generate hypotheses across large digital corpora.\n\n**Remote laboratories.** Facilities such as Emerald Cloud Lab provide genuine remote experimental capability rather than merely software simulation.\n\n**One-off experimental services.** Small projects do not necessarily require building a permanent laboratory relationship.\n\n**CRO and specialist supplier networks.** A large part of experimental science can already be purchased from external providers.\n\n**Research sourcing and contracting.** Platforms such as Science Exchange reduce some of the friction around finding and engaging qualified suppliers.\n\n**Digital execution records.** Automated and remote laboratories can preserve much richer operational records than a traditional handwritten lab notebook.\n\n**Predictable end-to-end pricing for small teams.** A nominal assay price is less useful when sample preparation, materials, controls and analysis appear later as separate costs.\n\n**Translation from AI proposal to executable protocol.** Someone still needs to turn scientific intention into a specification a real facility can run without changing the meaning of the experiment.\n\n**Comparable provider quotes.** Different laboratories often package scope, turnaround and deliverables differently, which makes comparison harder than choosing compute instances.\n\n**Visible scheduling.** Availability can determine whether an experiment is commercially useful, yet laboratory capacity is rarely exposed with the transparency we expect from cloud infrastructure.\n\n**Portable experimental records.** A result becomes much more valuable when another qualified provider can understand exactly how it was produced.\n\n**Reproducibility across providers.** The real test of a research marketplace is not only whether one laboratory can generate a number, but whether the work is documented well enough for somebody else to challenge or reproduce it.\n\nNotice that I am calling these **still to establish**, rather than **still missing**, because parts of this experience may already be available for particular techniques and providers. The open question is whether those pieces can become a sufficiently general, predictable workflow for small teams.\n\nThere is also a DeSci angle that I find considerably more convincing than “put research on-chain.”\n\nImagine a community wants to fund **one independent replication** of a published result. The group agrees on the question, selects a qualified provider, commits the budget and defines what evidence must be delivered: protocol version, raw data, deviations, instrument outputs and final report.\n\nA smart contract or other programmable funding mechanism could coordinate milestone payments and preserve references to the experimental record, while the scientific judgment would still come from researchers and qualified laboratories.\n\nThat division of labour matters because blockchain can help answer **who funded what, when the agreed work was delivered, and which record belongs to which experiment**; it cannot decide whether the biological conclusion is correct.\n\nUsed that way, Web3 becomes financial and provenance infrastructure around research rather than a substitute for research.\n\nCloud computing transformed software partly because a small company stopped needing to own every machine required by its product. AI is now doing something similar to parts of knowledge work because a small team can temporarily access reasoning capacity that would once have required a much larger group of specialists.\n\nExperimental science cannot become equally frictionless because atoms, samples, safety and instruments refuse to behave like virtual machines, although that does not mean the purchasing experience has to remain fragmented.\n\nIf AI can increasingly help us reach a useful experimental question, while remote labs and research marketplaces can already execute substantial portions of the physical work, then the interesting opportunity sits between them: **a trusted layer that turns a proposal into a reviewed protocol, a realistic price, an available laboratory and a reproducible result.**\n\nOnce that bridge becomes ordinary, a small research team may no longer need to own every instrument required by every question it wants to ask.\n\nAnd that is what I am thirsty for: not an AI that claims it has finished the science because it produced a convincing hypothesis, but an infrastructure layer that helps us discover what the physical world actually says.\n\n**Sources:**[Anthropic — Claude discovers novel enzyme system](https://www.anthropic.com/news/claude-discovers-novel-enzyme-system?utm_source=chatgpt.com)[Emerald Cloud Lab](https://www.emeraldcloudlab.com/?utm_source=chatgpt.com)[Science Exchange — supplier network](https://www.scienceexchange.com/suppliers?utm_source=chatgpt.com)[Science Exchange — for scientists](https://www.scienceexchange.com/biopharma/scientists?utm_source=chatgpt.com)\n\n[THIRSTDAY #7: WHERE CAN I ORDER THE EXPERIMENT MY AI JUST PROPOSED?](https://pub.towardsai.net/thirstday-7-where-can-i-order-the-experiment-my-ai-just-proposed-10cb978782d7) was originally published in [Towards AI](https://pub.towardsai.net) on Medium, where people are continuing the conversation by highlighting and responding to this story.", "url": "https://wpnews.pro/news/thirstday-7-where-can-i-order-the-experiment-my-ai-just-proposed", "canonical_source": "https://pub.towardsai.net/thirstday-7-where-can-i-order-the-experiment-my-ai-just-proposed-10cb978782d7?source=rss----98111c9905da---4", "published_at": "2026-10-02 06:40:38+00:00", "updated_at": "2026-10-02 07:07:46.410973+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-agents", "ai-research", "ai-safety"], "entities": ["Anthropic", "Claude", "Emerald Cloud Lab", "Science Exchange"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/thirstday-7-where-can-i-order-the-experiment-my-ai-just-proposed", "markdown": "https://wpnews.pro/news/thirstday-7-where-can-i-order-the-experiment-my-ai-just-proposed.md", "text": "https://wpnews.pro/news/thirstday-7-where-can-i-order-the-experiment-my-ai-just-proposed.txt", "jsonld": "https://wpnews.pro/news/thirstday-7-where-can-i-order-the-experiment-my-ai-just-proposed.jsonld"}}