Novo Nordisk said on May 19, 2026, that it has deployed an internal Azure-based reasoning agent built with Microsoft to help researchers run exploratory clinical analyses in minutes rather than weeks across its governed research data estate. The claim is not that a chatbot has discovered the next blockbuster medicine. It is that a large pharmaceutical company has found a practical place for agentic AI inside one of the slowest, most expensive, and most judgment-heavy parts of drug development. That distinction matters, because the most interesting enterprise AI stories are increasingly less about replacing experts and more about changing which questions experts can afford to ask.
The usual enterprise AI demo begins with a prompt and ends with a flourish. A user asks a question, the machine replies, and the audience is invited to imagine a future in which friction has been abolished. Novo Nordisk’s Azure agent is more prosaic than that, which is precisely why it is more important.
The company is not presenting the system as a general-purpose oracle for drug discovery. It is a quantitative decision-support agent aimed at pharmaceutical R&D, built to let researchers and medical experts explore hypotheses against proprietary clinical datasets using AI-generated code, statistical workflows, and human validation. In other words, the system lives where the real work lives: between a scientific hunch and the expensive decision to pursue it.
That is a useful corrective to the noisier version of the agent story. In regulated industries, the bottleneck is rarely that employees cannot type natural-language questions. The bottleneck is that data access, statistical rigor, compliance, review, and organizational trust all have to survive contact with production systems. Novo Nordisk’s project is interesting because it treats those constraints not as obstacles to AI adoption, but as the adoption problem itself.
Microsoft, naturally, would like this to be read as a validation of Azure as the preferred operating environment for enterprise agents. That is fair as far as it goes. But the more durable lesson is not “put an LLM near your data.” It is almost the opposite: if an AI system is going to influence high-value scientific choices, the hard part is everything that surrounds the model.
Novo Nordisk says the platform is aligned with industry standards including CDISC, SDTM, and ADaM. Those acronyms are not marketing decoration. They represent the kind of standardization that lets clinical data move from archival value to analytical utility, reducing the bespoke preparation work that so often turns a promising question into a weeks-long queue.
This is the part of the story that should make IT leaders slightly uncomfortable. Many organizations want the agent without the substrate. They want a natural-language interface to decades of institutional knowledge, but they have not resolved the identity model, the data quality problem, the lineage problem, or the governance model that would make such an interface safe.
Novo Nordisk’s example suggests that the winners in enterprise AI may not be the organizations with the most aggressive pilots. They may be the organizations whose boring infrastructure choices made the dramatic demo possible. The agent is the visible surface. The data estate is the leverage.
Novo Nordisk frames the Azure agent as a way to move from “gut-feel decision-making” toward quantitative decision support. That phrase deserves attention because it does not claim the system replaces scientific judgment. It claims the system changes the information environment in which judgment happens.
In practice, a researcher might ask whether a patient subgroup showed a particular treatment response, or whether a hypothesis is worth deeper investigation. Previously, answering that kind of question could require coordination across medical experts, commercial stakeholders, project managers, data scientists, and biostatisticians. The work was not just intellectually demanding; it was operationally expensive.
The promise of the agent is not that every exploratory answer becomes definitive. It is that more hypotheses can be evaluated earlier, weaker ideas can be discarded faster, and stronger ideas can be escalated with better supporting evidence. Novo Nordisk says teams that once had capacity to pursue perhaps 5–10 strong ideas in a quarter can now evaluate more than 50. Even if that number reflects early optimism, it points to a real shift in the economics of asking questions.
The danger, of course, is that faster analysis can create faster overconfidence. Exploratory work is not confirmatory evidence. A governed agent can draft analysis and surface patterns, but the clinical meaning of those patterns still requires expert interpretation. Novo Nordisk appears to understand that boundary, which is why the project’s emphasis on validation is more than a compliance footnote.
The agent can generate and execute code, work across structured clinical datasets, and produce explainable outputs for exploratory reasoning. That makes it closer to a tireless junior analyst than a digital scientist. It can reduce the waiting time between a question and an initial readout, but it does not own the final interpretation, the risk appetite, or the scientific accountability.
This distinction matters because enterprise AI often fails when organizations confuse interface with authority. A model that can produce a fluent answer is not necessarily entitled to influence a high-stakes decision. A system that can reason through statistical workflows still needs guardrails around what it can access, what it can execute, what it can claim, and who signs off before its output affects the real world.
Novo Nordisk says biostatisticians and subject-matter experts remain in the loop, validating promising outputs before they influence major scientific decisions. That is not a half measure. It is the only plausible way to deploy this class of system inside pharmaceutical R&D without turning productivity gains into governance debt.
The more subtle workforce story is also worth noting. When companies talk about AI augmentation, employees often hear a prelude to automation. Novo Nordisk’s framing is different: the goal is to remove low-value exploratory burden so scarce specialists can focus on the highest-priority opportunities. Whether that promise holds over time will depend on management incentives, not just technical design.
That is strategically important. The enterprise AI market is no longer impressed by a chatbot that summarizes a document. Customers want systems that can act inside workflows, respect access controls, produce auditable outputs, and justify their presence in budget meetings. Microsoft’s advantage is that it can pitch AI not as a separate product but as an extension of cloud, identity, security, developer tooling, and data services.
The Novo Nordisk project also fits a familiar Microsoft playbook. Rather than selling only a model endpoint, Microsoft embeds engineering help, reference architectures, and governance practices into customer deployments. The company’s Forward Deployed Engineering and AI Acceleration Studio language sounds like the enterprise version of a startup strike team: get close to the customer, wire the system into the real workflow, and turn the result into a repeatable pattern.
That does not mean Azure automatically wins the agent market. Pharmaceutical companies, banks, manufacturers, and public-sector organizations will all have multi-cloud estates, existing analytics stacks, and vendor risk concerns. But Microsoft’s strongest argument is not that it has the only capable models. It is that the model is only one layer of the system, and the rest of the system looks a lot like Microsoft’s home turf.
For WindowsForum readers, this is the larger platform implication. The AI contest is increasingly an infrastructure contest. The same identity, logging, policy, endpoint, compliance, and developer workflows that IT departments already manage are becoming the rails on which enterprise agents either scale or stall.
Novo Nordisk and Microsoft say they ran thousands of automated tests before production rollout and implemented layered evaluation systems covering query validation, system performance metrics, expert review, and continuous user feedback. That matters because agentic systems are not static applications. They interpret requests, select tools, generate code, and produce outputs whose failure modes can be more varied than a traditional dashboard.
The governance problem is not just hallucination. It is whether the system retrieved the right data, respected the right permissions, selected a defensible statistical approach, handled confounders appropriately, and presented uncertainty without laundering exploratory findings into false certainty. In clinical research, a beautifully formatted wrong answer is worse than a slow answer.
This is where many enterprise AI pilots quietly die. A system works in a controlled demo, then collapses under the weight of access control, auditability, exception handling, and organizational skepticism. Novo Nordisk’s experience suggests that skepticism was real at the start; the reported shift from “this is impossible” to “everyone wants access” is exactly the adoption curve vendors love to quote. But trust at scale is not a vibe. It has to be maintained through process.
The key question for Novo Nordisk now is whether governance can scale with demand. A handful of expert-supervised use cases is one thing. A broad internal rollout across trial design optimization, portfolio intelligence, predictive modeling, and multimodal data is another. The more capable the system becomes, the more important it is to define where its authority stops.
But elapsed time is not the same as decision quality. The deeper business claim is that the company can evaluate many more scientific questions earlier, improving the probability that the right ideas advance and the wrong ideas stop sooner. That is both more valuable and harder to prove.
In drug development, probability of success is influenced by biology, trial design, patient selection, regulatory context, commercial strategy, and plain uncertainty. An agent that improves early analysis may help, but attributing downstream trial success to that agent will be difficult. The honest case for the system is not that it guarantees better outcomes. It is that it increases the number and quality of informed decisions made before the most expensive commitments.
That distinction should guide how other organizations assess similar tools. The first metrics are likely to be operational: time to first analysis, number of hypotheses evaluated, biostatistician hours redirected, user adoption, error rates, and review outcomes. The harder metrics come later: better portfolio prioritization, fewer late-stage surprises, improved trial design, and higher R&D productivity.
The risk is that the speed metric becomes the headline because it is easy to understand. Weeks to minutes is compelling. But Novo Nordisk’s more consequential claim is about expanding the search space of scientific inquiry while preserving rigor. Speed is the accelerator; judgment is still the steering wheel.
Descriptive analysis asks what the data shows. Predictive modeling asks what may happen next. Simulation asks how different choices could alter the outcome. Each step moves the system closer to strategic decision-making, and each step increases the risk that users treat model output as more settled than it is.
The company also wants the agent to reason across broader data modalities, including omics, device telemetry, and external scientific literature. That expansion makes sense scientifically. Modern drug development increasingly depends on connecting clinical outcomes with biological signals, real-world evidence, monitoring data, and the published research landscape.
Technically and operationally, however, multimodal reasoning is a messier world. Different data types have different quality profiles, consent constraints, privacy implications, and analytical assumptions. Literature may be outdated or contradictory. Device telemetry may be noisy. Omics data can be high-dimensional and easy to overinterpret. The more sources an agent can synthesize, the more carefully it must show its work.
This is why Novo Nordisk’s initial emphasis on structured clinical data is significant. It starts the agent in a domain where the data foundation is comparatively mature and the validation practices are well understood. The roadmap is ambitious, but the deployment sequence suggests a company trying to earn its way into more complex territory rather than sprinting there for a press release.
That is why the phrase “agent” is both useful and misleading. The market often treats agents as autonomous workers. In practice, the most viable enterprise agents may look more like governed accelerators for expert workflows. They do not replace the accountable professional; they reduce the cost of preparing, testing, and documenting options before the professional decides.
For IT leaders, the lesson is to stop asking whether the model is smart enough in the abstract. The better question is whether the organization can define a bounded workflow where AI-generated work can be evaluated, audited, and improved. Novo Nordisk chose exploratory quantitative analysis because it had the data foundation, the expert validators, and a clear bottleneck.
That pattern is portable. A bank might use similar systems for risk scenario analysis. A manufacturer might use them for quality investigations. A security team might use them for incident triage. The shared requirement is not magical autonomy; it is controlled reasoning over trusted data with human review at the decision boundary.
The companies that succeed will likely be those that resist two temptations. The first is deploying AI too broadly, producing shallow assistants that impress briefly and then fade. The second is deploying AI too timidly, trapping it in pilots that never touch meaningful work. Novo Nordisk’s case sits in the more interesting middle: narrow enough to govern, important enough to matter.
Novo Nordisk Turns the AI Pitch Into a Workflow Problem
The usual enterprise AI demo begins with a prompt and ends with a flourish. A user asks a question, the machine replies, and the audience is invited to imagine a future in which friction has been abolished. Novo Nordisk’s Azure agent is more prosaic than that, which is precisely why it is more important.The company is not presenting the system as a general-purpose oracle for drug discovery. It is a quantitative decision-support agent aimed at pharmaceutical R&D, built to let researchers and medical experts explore hypotheses against proprietary clinical datasets using AI-generated code, statistical workflows, and human validation. In other words, the system lives where the real work lives: between a scientific hunch and the expensive decision to pursue it.
That is a useful corrective to the noisier version of the agent story. In regulated industries, the bottleneck is rarely that employees cannot type natural-language questions. The bottleneck is that data access, statistical rigor, compliance, review, and organizational trust all have to survive contact with production systems. Novo Nordisk’s project is interesting because it treats those constraints not as obstacles to AI adoption, but as the adoption problem itself.
Microsoft, naturally, would like this to be read as a validation of Azure as the preferred operating environment for enterprise agents. That is fair as far as it goes. But the more durable lesson is not “put an LLM near your data.” It is almost the opposite: if an AI system is going to influence high-value scientific choices, the hard part is everything that surrounds the model.
The Real Asset Was Not the Model, but the Data Foundation
The most revealing part of Novo Nordisk’s story is that the agent arrived after years of data groundwork. The company’s FounData initiative reportedly harmonized more than 200,000 patient-years of clinical trial data across studies, disease areas, and research programs. That foundation matters because clinical AI systems do not become useful merely by being fluent; they become useful when they can operate against data that has already been cleaned, structured, governed, and made analytically meaningful.Novo Nordisk says the platform is aligned with industry standards including CDISC, SDTM, and ADaM. Those acronyms are not marketing decoration. They represent the kind of standardization that lets clinical data move from archival value to analytical utility, reducing the bespoke preparation work that so often turns a promising question into a weeks-long queue.
This is the part of the story that should make IT leaders slightly uncomfortable. Many organizations want the agent without the substrate. They want a natural-language interface to decades of institutional knowledge, but they have not resolved the identity model, the data quality problem, the lineage problem, or the governance model that would make such an interface safe.
Novo Nordisk’s example suggests that the winners in enterprise AI may not be the organizations with the most aggressive pilots. They may be the organizations whose boring infrastructure choices made the dramatic demo possible. The agent is the visible surface. The data estate is the leverage.
From Gut Feel to Quantitative Triage
Pharmaceutical R&D is a brutal filtering machine. A company can spend years and enormous sums moving a candidate through trials, only to discover that the evidence does not support the hoped-for outcome. That does not make failure avoidable, but it makes earlier and better triage immensely valuable.Novo Nordisk frames the Azure agent as a way to move from “gut-feel decision-making” toward quantitative decision support. That phrase deserves attention because it does not claim the system replaces scientific judgment. It claims the system changes the information environment in which judgment happens.
In practice, a researcher might ask whether a patient subgroup showed a particular treatment response, or whether a hypothesis is worth deeper investigation. Previously, answering that kind of question could require coordination across medical experts, commercial stakeholders, project managers, data scientists, and biostatisticians. The work was not just intellectually demanding; it was operationally expensive.
The promise of the agent is not that every exploratory answer becomes definitive. It is that more hypotheses can be evaluated earlier, weaker ideas can be discarded faster, and stronger ideas can be escalated with better supporting evidence. Novo Nordisk says teams that once had capacity to pursue perhaps 5–10 strong ideas in a quarter can now evaluate more than 50. Even if that number reflects early optimism, it points to a real shift in the economics of asking questions.
The danger, of course, is that faster analysis can create faster overconfidence. Exploratory work is not confirmatory evidence. A governed agent can draft analysis and surface patterns, but the clinical meaning of those patterns still requires expert interpretation. Novo Nordisk appears to understand that boundary, which is why the project’s emphasis on validation is more than a compliance footnote.
The Agent Is a Junior Analyst, Not a Digital Scientist
There is a reason Novo Nordisk’s spokespeople keep returning to the same point: the system informs decisions; it does not make them. That may sound like cautious corporate language, but in this context it is the architectural principle.The agent can generate and execute code, work across structured clinical datasets, and produce explainable outputs for exploratory reasoning. That makes it closer to a tireless junior analyst than a digital scientist. It can reduce the waiting time between a question and an initial readout, but it does not own the final interpretation, the risk appetite, or the scientific accountability.
This distinction matters because enterprise AI often fails when organizations confuse interface with authority. A model that can produce a fluent answer is not necessarily entitled to influence a high-stakes decision. A system that can reason through statistical workflows still needs guardrails around what it can access, what it can execute, what it can claim, and who signs off before its output affects the real world.
Novo Nordisk says biostatisticians and subject-matter experts remain in the loop, validating promising outputs before they influence major scientific decisions. That is not a half measure. It is the only plausible way to deploy this class of system inside pharmaceutical R&D without turning productivity gains into governance debt.
The more subtle workforce story is also worth noting. When companies talk about AI augmentation, employees often hear a prelude to automation. Novo Nordisk’s framing is different: the goal is to remove low-value exploratory burden so scarce specialists can focus on the highest-priority opportunities. Whether that promise holds over time will depend on management incentives, not just technical design.
Microsoft Gets the Enterprise Agent Story It Wanted
For Microsoft, the Novo Nordisk case lands neatly inside a broader Azure narrative. The company has spent the past two years trying to move customers from AI experimentation into production agents, with Azure AI Foundry, Azure OpenAI Service, orchestration tools, evaluation systems, and governance patterns all positioned as part of the same enterprise platform story. Novo Nordisk gives Microsoft something more persuasive than a keynote demo: a regulated customer using agents against valuable proprietary data.That is strategically important. The enterprise AI market is no longer impressed by a chatbot that summarizes a document. Customers want systems that can act inside workflows, respect access controls, produce auditable outputs, and justify their presence in budget meetings. Microsoft’s advantage is that it can pitch AI not as a separate product but as an extension of cloud, identity, security, developer tooling, and data services.
The Novo Nordisk project also fits a familiar Microsoft playbook. Rather than selling only a model endpoint, Microsoft embeds engineering help, reference architectures, and governance practices into customer deployments. The company’s Forward Deployed Engineering and AI Acceleration Studio language sounds like the enterprise version of a startup strike team: get close to the customer, wire the system into the real workflow, and turn the result into a repeatable pattern.
That does not mean Azure automatically wins the agent market. Pharmaceutical companies, banks, manufacturers, and public-sector organizations will all have multi-cloud estates, existing analytics stacks, and vendor risk concerns. But Microsoft’s strongest argument is not that it has the only capable models. It is that the model is only one layer of the system, and the rest of the system looks a lot like Microsoft’s home turf.
For WindowsForum readers, this is the larger platform implication. The AI contest is increasingly an infrastructure contest. The same identity, logging, policy, endpoint, compliance, and developer workflows that IT departments already manage are becoming the rails on which enterprise agents either scale or stall.
Governance Becomes the Product Feature
The phrase governed reasoning agent sounds like something designed by committee, but it captures the crux of the deployment. In consumer AI, the magic is the answer. In regulated enterprise AI, the magic is being able to explain why the answer was allowed to exist.Novo Nordisk and Microsoft say they ran thousands of automated tests before production rollout and implemented layered evaluation systems covering query validation, system performance metrics, expert review, and continuous user feedback. That matters because agentic systems are not static applications. They interpret requests, select tools, generate code, and produce outputs whose failure modes can be more varied than a traditional dashboard.
The governance problem is not just hallucination. It is whether the system retrieved the right data, respected the right permissions, selected a defensible statistical approach, handled confounders appropriately, and presented uncertainty without laundering exploratory findings into false certainty. In clinical research, a beautifully formatted wrong answer is worse than a slow answer.
This is where many enterprise AI pilots quietly die. A system works in a controlled demo, then collapses under the weight of access control, auditability, exception handling, and organizational skepticism. Novo Nordisk’s experience suggests that skepticism was real at the start; the reported shift from “this is impossible” to “everyone wants access” is exactly the adoption curve vendors love to quote. But trust at scale is not a vibe. It has to be maintained through process.
The key question for Novo Nordisk now is whether governance can scale with demand. A handful of expert-supervised use cases is one thing. A broad internal rollout across trial design optimization, portfolio intelligence, predictive modeling, and multimodal data is another. The more capable the system becomes, the more important it is to define where its authority stops.
The Productivity Gain Is Real, but So Is the Measurement Problem
Novo Nordisk says the agent could reduce time to insight from weeks to minutes for many exploratory analyses. That is the kind of claim enterprise AI buyers want to hear, and it is plausible in a specific sense. If a researcher previously had to wait for a biostatistics queue to produce an initial exploratory analysis, and the agent can draft that analysis on demand, the elapsed time can collapse dramatically.But elapsed time is not the same as decision quality. The deeper business claim is that the company can evaluate many more scientific questions earlier, improving the probability that the right ideas advance and the wrong ideas stop sooner. That is both more valuable and harder to prove.
In drug development, probability of success is influenced by biology, trial design, patient selection, regulatory context, commercial strategy, and plain uncertainty. An agent that improves early analysis may help, but attributing downstream trial success to that agent will be difficult. The honest case for the system is not that it guarantees better outcomes. It is that it increases the number and quality of informed decisions made before the most expensive commitments.
That distinction should guide how other organizations assess similar tools. The first metrics are likely to be operational: time to first analysis, number of hypotheses evaluated, biostatistician hours redirected, user adoption, error rates, and review outcomes. The harder metrics come later: better portfolio prioritization, fewer late-stage surprises, improved trial design, and higher R&D productivity.
The risk is that the speed metric becomes the headline because it is easy to understand. Weeks to minutes is compelling. But Novo Nordisk’s more consequential claim is about expanding the search space of scientific inquiry while preserving rigor. Speed is the accelerator; judgment is still the steering wheel.
The Roadmap Moves From Describing Data to Simulating Futures
Novo Nordisk says it is looking beyond descriptive analyses toward predictive capabilities, including outcome modeling and trial-scenario simulation. That is the natural next step, and also the point at which the governance burden increases sharply.Descriptive analysis asks what the data shows. Predictive modeling asks what may happen next. Simulation asks how different choices could alter the outcome. Each step moves the system closer to strategic decision-making, and each step increases the risk that users treat model output as more settled than it is.
The company also wants the agent to reason across broader data modalities, including omics, device telemetry, and external scientific literature. That expansion makes sense scientifically. Modern drug development increasingly depends on connecting clinical outcomes with biological signals, real-world evidence, monitoring data, and the published research landscape.
Technically and operationally, however, multimodal reasoning is a messier world. Different data types have different quality profiles, consent constraints, privacy implications, and analytical assumptions. Literature may be outdated or contradictory. Device telemetry may be noisy. Omics data can be high-dimensional and easy to overinterpret. The more sources an agent can synthesize, the more carefully it must show its work.
This is why Novo Nordisk’s initial emphasis on structured clinical data is significant. It starts the agent in a domain where the data foundation is comparatively mature and the validation practices are well understood. The roadmap is ambitious, but the deployment sequence suggests a company trying to earn its way into more complex territory rather than sprinting there for a press release.
Regulated Industries Are Watching the Pattern, Not the Product
The broader significance of Novo Nordisk’s deployment is not limited to pharmaceutical R&D. Banks, insurers, manufacturers, energy companies, hospitals, and government agencies all have versions of the same problem: critical decisions depend on specialized experts, fragmented data, regulated processes, and expensive delays.That is why the phrase “agent” is both useful and misleading. The market often treats agents as autonomous workers. In practice, the most viable enterprise agents may look more like governed accelerators for expert workflows. They do not replace the accountable professional; they reduce the cost of preparing, testing, and documenting options before the professional decides.
For IT leaders, the lesson is to stop asking whether the model is smart enough in the abstract. The better question is whether the organization can define a bounded workflow where AI-generated work can be evaluated, audited, and improved. Novo Nordisk chose exploratory quantitative analysis because it had the data foundation, the expert validators, and a clear bottleneck.
That pattern is portable. A bank might use similar systems for risk scenario analysis. A manufacturer might use them for quality investigations. A security team might use them for incident triage. The shared requirement is not magical autonomy; it is controlled reasoning over trusted data with human review at the decision boundary.
The companies that succeed will likely be those that resist two temptations. The first is deploying AI too broadly, producing shallow assistants that impress briefly and then fade. The second is deploying AI too timidly, trapping it in pilots that never touch meaningful work. Novo Nordisk’s case sits in the more interesting middle: narrow enough to govern, important enough to matter.
The Novo Nordisk Case Gives IT a More Useful AI Scorecard
The practical take from Novo Nordisk’s Azure agent is not that every enterprise should copy its architecture tomorrow. It is that serious AI deployment has a recognizable shape: governed data, bounded workflows, measurable bottlenecks, expert review, and a plan for scaling trust rather than merely scaling access.- The agent’s value depends heavily on Novo Nordisk’s prior investment in harmonized clinical data, not just on the capabilities of the underlying AI model.
- The strongest near-term productivity claim is the reduction of exploratory analysis cycles from weeks to minutes for suitable questions.
- The system is positioned as an augmentation layer for researchers and biostatisticians, with human validation remaining central to scientific decision-making.
- Microsoft gains a credible enterprise proof point for Azure-based agents in a regulated, high-value workflow.
- The next phase, including predictive modeling and multimodal reasoning, will test whether the governance model can scale with the system’s ambition.
- Other regulated industries should study the deployment pattern more than the pharmaceutical specifics.
References
- Primary source: Microsoft
Published: 2026-05-19T20:30:08.524633
Novo Nordisk accelerates clinical insight with custom agents on Azure | Microsoft Customer Stories
Novo Nordisk accelerates clinical insight generation with a custom AI agent that helps deliver rapid, compliant data analysis.www.microsoft.com
- Official source: azure.microsoft.com
Transforming R&D with agentic AI: Introducing Microsoft Discovery | Microsoft Azure Blog
Learn more about the new enterprise agentic platform, , accelerating research and development
azure.microsoft.com
- Related coverage: mongodb.com
Novo Nordisk & MongoDB Atlas: Groundbreaking Time To Value Acceleration With A Clinical Study Report In Minutes
Danish pharmaceutical giant becomes the first in the industry to generate a complete Clinical Study Report in minutes with Generative AI And MongoDB Atlas.www.mongodb.com
- Related coverage: celonis.com
Novo Nordisk + Celonis
Learn how Novo Nordisk is speeding up clinical development by orchestrating 100 AI agents through Celonis. Their vision: reduce time-to-market by 12 months.www.celonis.com - Official source: cdn-dynmedia-1.microsoft.com
- Official source: techcommunity.microsoft.com
- Official source: adoption.microsoft.com
- Official source: marketingassets.microsoft.com
