Fast Brain vs. Smart Brain in Life Sciences: What every Gen AI User Needs to Know

Hey Friends,

If you've been keeping up with AI news at all since February, you've probably heard people talking about agents. OpenClaw—the open-source AI agent that went from obscure side project to 247,000 GitHub stars in a matter of weeks—has been the latest viral craze. Even I haven't been immune. I spent the last several days building three agent buddies of my own.

But forget about that. Seriously.

If you haven't been sprinting alongside the AI train since 2023, it might feel like you've been left in the dust. I get it. None of this has ever been what I'd call intuitive. The terminology changes constantly, new tools launch every week, and the loudest voices online are usually the ones building things most of us will never need.

So let's slow down and talk about what you actually need to know.

I'm a huge fan of Professor Ethan Mollick's writing, and every few months he does a superb job of encapsulating "how to use AI right now." His latest post—A Guide to Which AI to Use in the Agentic Era—is an excellent expert-level summary of today's AI models, tools, and vocab. If you want the full landscape, I strongly suggest reading it.

But if, like many of my readers, you're brand new to this—and beyond knowing that AI literacy is important, you're not sure where to start—here's what I think matters most for those of us in life sciences.

First off, most people work in a chat app. That's ChatGPT, Claude, Gemini, or an enterprise equivalent. That's the interface. But there are a few things happening under the hood that you need to understand:

1. The AI model is the brain—and it can be fast or it can be smart (not both -yet).

There are faster models optimized for casual chat, and then there are reasoning (or "thinking") models built for accuracy and complex problem-solving. Unless you're just having a fun conversation where being right doesn't matter, use a thinking model. Use it when you're trying to understand something complex. Use it when the output is going into your work. Use it when accuracy matters.

In CMC, QA, and regulatory roles? That's all the time.

2. Understand what tools your model has access to.

The ability to browse, search, and read web content is a tool—and your AI model can't do anything on the internet unless that tool is turned on. Most people I talk to are surprised to hear that their enterprise chatbot (the one approved for work) often does not have web search enabled for security reasons. That means it can't give you specific, real-time answers. If you're asking it about the latest FDA guidance or a list of products in phase 2 trials, and it doesn't have web access, it's guessing from training data—or worse, making something up confidently.

3. If you can't select your model, assume you're getting the dumber one.

If you're using an app that doesn't let you manually choose a reasoning or thinking model (cough cough, Copilot…), you should assume that what you're getting is fast but less capable. That matters. The difference between a default model and a thinking model on a regulatory question can be the difference between a useful draft and a confidently wrong one. Same goes for free tier ChatGPT and others. Good models cost more money to run relatively speaking— and at the moment— you very much get what you pay for.

What I’m trying to say is, take AI literacy seriously. It hasn't gotten any easier to figure out how to use these tools. If anything, it's gotten harder. But you don't want to miss out on the most powerful tools available today just because it's hard to know the difference between them. Start with the basics: know your model, know your tools, and know when to demand accuracy over speed.

The learning curve is real. But so is the payoff.

Thanks for reading!

-Alexa

News

EMA and Industry Sit Down to Talk AI—Here's What Was Said

What's New: On February 4, 2026, the European Medicines Agency (EMA) and the Heads of Medicines Agencies (HMA) held their latest AI-focused meeting with industry stakeholders. The agenda covered a lot of ground: the newly published EMA-FDA joint AI guiding principles, upcoming AI guidance across clinical development, pharmacovigilance, and GMP manufacturing (Annex 22), and the implications of the proposed EU Biotech Act—which would make it a legal requirement for EMA to develop AI guidance across the entire drug lifecycle. Industry groups, including EFPIA, EUCOPE, EuropaBio, and others, came with a clear message: we need practical, operational guidance—not just high-level principles.

How It Works:

• EMA-FDA AI Guiding Principles: Published in January 2026, these ten jointly developed principles establish a shared transatlantic framework for AI use across the medicines lifecycle—covering nonclinical research, clinical trials, manufacturing, and post-market surveillance. They emphasize a risk-based, human-centric approach with strong data governance and transparency. A glossary of harmonized AI terminology is currently being developed between the two agencies.

• Guidance Pipeline: EMA has several AI guidance efforts in the works: a concept paper on AI in clinical development, a Q&A-style guidance on AI in pharmacovigilance (developed with PRAC), and the high-profile GMP Annex 22 on AI in manufacturing, which received roughly 1,300 public comments during consultation and is expected to be finalized by end of 2026.

• Biotech Act: The proposed EU Biotech Act would legally require EMA to develop and publish AI guidance across pre-clinical, clinical, manufacturing, post-authorization, and regulatory authorization phases—in coordination with the European Commission and AI Office. EMA signaled that work will begin before the Act is formally adopted.

• Pharmacovigilance Use Cases: The meeting highlighted seven emerging AI use cases within Europe's signal detection workflow, including AI-powered literature screening for safety reports, triaging serious adverse events, detecting drug reactions in real-world data, and AI-enhanced case adjudication. This was framed as one of the most promising areas for near-term AI adoption.

• Industry's Wishlist: Industry associations asked for checklists and templates (not just principles), risk-based frameworks that differentiate high vs. low-risk AI use cases, clarity on acceptable uses of generative AI in regulatory submissions, transparency expectations for AI-generated content, structured regulatory data access via APIs, and a more flexible approach to using LLMs and generative AI in GMP applications.

• Regulators' Response: EMA was candid: they are still in "listening mode," there is no official regulatory position yet on most of the points industry raised, and detailed prescriptive rules may not be feasible given how context-dependent AI use cases are. Responsibility for compliance still sits with sponsors and applicants.

Why It Matters: This meeting is a clear signal that the regulatory conversation around AI in life sciences is accelerating—and getting more specific. For anyone working in CMC, QA, or regulatory affairs, several things stand out:

GMP Annex 22 is coming. The current draft applies to static, deterministic AI models in critical GMP applications and explicitly excludes generative AI and LLMs from GMP-critical decisions. Industry is pushing for more flexibility here, but the draft's framework around validation, explainability, and data independence reflects how regulators expect AI to be governed in manufacturing environments.

The Biotech Act could change the game. If adopted, it transforms AI guidance from a "nice to have" into a legal mandate—meaning regulatory expectations will formalize faster than many expect.

Pharmacovigilance is where AI adoption may land first. Shared processes between industry and regulators, combined with clear efficiency gains, make PV a natural proving ground.

And perhaps most importantly: regulators are telling industry, "We're building the framework, but you own the compliance." No amount of guidance will remove the obligation to validate, document, and justify how AI is used in your regulated workflows.

My Take: I've said before that the EMA came into the AI conversation too rigid. Their 2024 reflection paper set a cautious tone—generative language models were called out specifically as hallucination-prone, AI-generated text for regulatory documents required "close human supervision," and the overall message felt like a strong discouragement of gen AI for anything regulatory-adjacent, whether you had a human in the loop or not. Meanwhile, the FDA's January 2025 draft guidance explicitly scoped out internal workflows and document drafting—essentially saying, "if it's not generating evidence for a regulatory decision, it's not really our concern." That contrast was stark: FDA gave the industry room to experiment with AI for efficiency, while EMA seemed to be saying "proceed with extreme caution, if at all."

This February meeting looks like a walkback of some of that rigidity—or at least the beginning of one. Industry associations came in loud and clear asking for regulatory clarity on acceptable uses of generative AI in submissions, transparency expectations when AI-generated content is used, and—notably—a more flexible approach to using LLMs and generative AI in GMP applications. And instead of shutting that down, the EMA said they're in "listening mode" and that industry's input would feed into scoping discussions for upcoming guidance. That's a meaningful shift in posture. They're no longer just warning about the risks of gen AI—they're acknowledging that industry is already using it and that the guidance needs to meet reality where it is.

The joint EMA-FDA principles published in January also help here. By aligning with the FDA on a shared, risk-based, principles-first framework, the EMA has given itself room to move away from the more prescriptive stance of the reflection paper without looking like it's lowering the bar. It's not a green light for generative AI in GxP workflows—not even close. But the door that felt mostly shut in 2024 is now cracked open, and the conversation has shifted from "should we allow this?" to "how do we govern this responsibly?" For those of us in life sciences, that's progress. Start documenting how you use AI now, because the compliance framework is forming in real time—and this meeting tells me it's going to be more pragmatic than I originally expected.

Source: Summary notes of the HMA-EMA group focused on AI with industry stakeholders meeting — February 4, 2026

ASCO Launches a Dedicated AI-in-Oncology Platform

What's New: On February 10, the American Society of Clinical Oncology (ASCO) and digital health publisher Conexiant launched ASCO AI in Oncology (ascoai.org)—a dedicated digital platform designed to help oncology professionals make sense of AI's rapidly expanding role in cancer care. This is the second major collaboration between ASCO and Conexiant, following the long-running success of The ASCO Post, which has been a go-to oncology news resource since 2010. The platform is already live with articles, expert commentary, and video content.

How It Works: The platform is organized around three content pillars: Trusted Intelligence (curated, expert-led insights prioritizing clinical relevance), Practical Education (focused on how AI tools can enhance diagnostic accuracy, streamline workflows, and personalize treatment), and Community Perspective (a space for clinicians and researchers to share real-world implementation experiences across disciplines and institutions).

Content already on the site includes clinical deep dives—like an AI risk prediction model for cardiovascular events in cancer patients, and a CNN-based tool matching dermatologists in grading cutaneous squamous cell carcinomas—alongside practitioner-submitted commentaries, including a patient-perspective piece on using AI tools to inform treatment decisions for multiple myeloma.

The platform also links directly to ASCO's existing AI infrastructure: the AI-powered ASCO Guidelines Assistant (built in collaboration with Google Cloud and Wolters Kluwer), ASCO's Guiding Principles for Responsible AI Use in Oncology (first published in 2024 and recently updated), and a glossary of AI terminology. Additional functionality, including patient-focused content, is planned for the coming months.

This isn't ASCO's first move in the AI space—it builds on a broader institutional strategy that includes embedding AI literacy into their meetings and educational programming—but it's the first time they've stood up a dedicated, always-on platform for it.

Why It Matters: ASCO represents over 50,000 oncology professionals globally, and when an organization of that size and influence launches a dedicated AI platform, it sends a signal to the broader life sciences community: AI literacy isn't optional anymore, even in clinical specialties.

For those of us working in CMC, QA, and regulatory roles, the relevance might seem indirect—oncology clinical care is a different world from batch records and submission dossiers. But the underlying problem ASCO is trying to solve is the same one we face: AI is moving faster than the shared understanding of how to use it. ASCO CEO Clifford Hudis put it bluntly—there's too much noise in the AI space, and clinicians need a trusted filter. That's true whether you're an oncologist evaluating a diagnostic AI tool or a regulatory professional trying to figure out what gen AI use is acceptable in a submission workflow.

The platform also reflects a pattern worth watching: major professional societies are shifting from publishing occasional position papers on AI to building persistent, living resources. That's an acknowledgment that AI isn't a one-time policy question—it's an ongoing literacy challenge.

My Take: What I like about this is that ASCO isn't just talking about AI—they're building the infrastructure for their members to learn about it in a structured, credible way. That's the piece most professional communities in life sciences are still missing. We have guidance documents and reflection papers, but we don't have many dedicated, curated spaces where practitioners can go to consistently learn what's real, what's hype, and what's actually ready for their workflows. The fact that ASCO is doing this in partnership with an established health publishing platform (rather than building a chatbot or launching a tool) tells you something about where the real need is right now: it's not more AI products, it's more AI understanding. I'd love to see ISPE, PDA, or RAPS build something equivalent for the CMC and regulatory community. The demand is there—someone just needs to build the trusted home for it.

Source: ASCO and Conexiant to Launch ASCO AI in Oncology — February 10, 2026