AI Weekly’s directory lists 28 named AI deployments in research and development, spanning software, biotech, scientific research, manufacturing, transportation and healthcare. The count describes cases the directory included—not 28 independently verified successes. Its status labels mix production systems, reported results, pilots and efforts described as halted or reversed, so the roundup is best read as a map of applications and claims, not a comparable scorecard.
What does “28 real deployments” count?
AI Weekly says its directory contains 28 named deployments and was last updated September 28, 2026. “Real” is the directory’s framing: the number counts entries in that roundup. It does not establish that every system is currently operating, that each entry reached production, or that an independent reviewer confirmed each account.
The directory reports 17 deployments “in production or with results,” 17 “with a reported outcome,” and four “halted or reversed.” These are directory classifications, not results from a common audit. The categories may overlap and should not be added together as though they divide the 28 cases into mutually exclusive groups. The directory’s summary does not make the underlying status or outcome evidence comparable across entries.
Which industries and R&D tasks appear?
The roundup covers work from research support to operational systems. Its sector counts are the directory’s own classifications:
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →#1 Best Overall
| Directory category | Entries | Examples of work represented |
|---|---|---|
| Software & Tech | 11 | Internal research agents and model development |
| Pharma & Biotech | 9 | Molecule and drug discovery, laboratory biology |
| Science & Research | 5 | Scientific hypothesis generation and research workflows |
| Manufacturing | 1 | Semiconductor simulation and design |
| Transportation | 1 | Autonomous-vehicle training data |
| Healthcare | 1 | Clinical-trial screening |
AI Weekly names NaiveAI, OpenAI, Anthropic and Hugging Face in Software & Tech. In Pharma & Biotech, it names Enveda, Novo Nordisk, Anew Labs, Isomorphic Labs, Anthropic, Gamgee, Eli Lilly, Amgen, Moderna, Allen Institute and Thermo Fisher. Its Science & Research group names Google, Anthropic, Fermi Explorer Mission and the U.S. Department of Energy National Laboratories; the remaining sector entries name Intel, Uber and Cleveland Clinic, respectively. These are names reported by the directory, not independently verified deployment descriptions here. The names should not be read as a one-to-one count of entries: for example, the directory reports nine Pharma & Biotech entries while naming more organizations.
How AI is being used across the R&D process
R&D is not one task. The cases described in the roundup range from tools that help people find and synthesize information to systems aimed at generating scientific or operational outputs. That distinction matters: a research assistant can accelerate a workflow without itself demonstrating a better drug, discovery or manufacturing result.
Research workflow support
Some examples concern internal research agents, model development, information synthesis or hypothesis generation. These are ways to help researchers navigate knowledge, formulate questions or work with data. An announced tool or pilot in this category is evidence that an organization is exploring a workflow; it is not, on its own, evidence that a scientific finding has been validated or that the tool improves research outcomes.
Drug and biological discovery
Drug-development applications can target different stages: identifying biological targets, proposing or selecting molecules, and informing later study design. Novartis describes its own AI-enabled R&D strategy as spanning those kinds of questions. The company says digital technologies, many powered by AI, help teams process large volumes of information; its examples include assessing promising biological targets, selecting molecules with fewer side effects, and combining generative AI with knowledge graphs to summarize prior studies and real-world evidence for clinical-trial design. This is Novartis’s account of its approach, not an independent performance assessment or proof that any particular candidate will succeed.
Novartis frames the aim as faster decision-making, while noting the resource demands and uncertainty of drug discovery. The company-published questions—“Which biological targets look most promising?” and “Which molecules might work on those targets with the fewest side-effects?”—capture the kinds of decisions AI may support. They do not establish that AI can answer them reliably without scientific review.
Scientific and operational applications
Other examples in the directory touch semiconductor simulation and design, autonomous-vehicle training data, and clinical-trial screening. These are materially different from molecule discovery: their outputs, constraints and ways to measure utility differ. The directory also includes laboratory biology and broader scientific research applications, but its summary does not provide enough case-level detail to assign specific methods, oversight arrangements or measured outcomes to every named organization.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What counts as a result—and how strong is the evidence?
A deployment label and an evidence claim answer different questions. “Pilot” indicates a trial or limited use rather than proof of durable production value; “in production” indicates reported operational use, not necessarily independently measured impact. “Reported outcome” means an outcome is described by the directory, but the label alone does not say who measured it, under what conditions, or whether it was independently reproduced. “Halted or reversed” signals that a case did not continue as originally described; it is not interchangeable with a failed scientific experiment.
For each case, a careful reader would want to know the task, stage, human oversight, provenance of the evidence, and the outcome actually measured. The directory’s aggregate counts do not supply a shared measure across cases, and its listed status categories should not be treated as a ranking of technical quality or scientific value.
Recommended Free Tools
AI Weekly is the source for the 28-case count, sector breakdown and status summaries. Novartis is the primary source for its own description of its R&D strategy. A peer-reviewed review of AI across drug development provides broader lifecycle context and company case studies, but it does not independently establish the details or current status of every directory entry. The available source material does not independently verify all 28 cases or provide comparable measurements for them; claims about specific organizations should therefore be attributed to the source making them rather than presented as audited facts.
Quick Recap
How to interpret the roundup
- Use it as a landscape, not a leaderboard. The cases span different scientific and operational tasks, so a single success metric would obscure important differences.
- Separate activity from impact. An announcement, pilot, production deployment and reported result are different kinds of evidence.
- Check who is making each claim. A company account can explain the intended use; independent validation is a separate question.
- Look for the actual outcome. A useful assessment identifies what changed, how it was measured, and whether people remained responsible for review and decisions.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
