🌿 THE GOOD AI
The blood test that could change pancreatic cancer forever
Pancreatic cancer is one of medicine's cruelest realities: by the time most patients get a diagnosis, there's almost nothing that can be done. Roughly 90% of patients die within five years, not because treatment doesn't exist, but because the cancer is almost never caught before it spreads.
That may finally be changing. Researchers at Taiwan's National Taiwan University Hospital built PanMETAI, an AI model that scans hundreds of metabolic signals in just half a millilitre of blood. In clinical trials, it achieved 94% diagnostic accuracy for early-stage pancreatic cancer, and those results held up when tested against a separate European validation cohort. This isn't a one-country lab result. It's a model that works across populations.
We want to be clear-eyed: there's a long road from clinical trial to widespread use. Regulatory approval, cost, and access infrastructure all stand between this result and a world where pancreatic cancer is routinely caught early. But the existence of a validated approach, peer-reviewed and published in Nature Communications, reorients what's possible. We now know it can be done.
The direction is right. And in medicine, direction matters.
⚡ 3 GOOD SIGNALS
🌿 Wildlife science just became open-source
Google has released SpeciesNet, trained on 65 million labelled camera-trap images, with 94.5% species identification accuracy - under a free-to-use licence. What previously took conservation teams weeks of manual tagging can now be processed in hours. Pumas in Colombia, lions in the Serengeti, cassowaries in Australia, SpeciesNet is already tracking them all. Alongside the release, Google is launching an accelerator and grant program for nature-focused AI startups. Open-sourcing infrastructure this powerful is what "tech for good" actually looks like.
📈 The economic case for AI just got harder to dismiss
For years, economists argued AI hadn't yet shown up in the macro data. A major study of 12,000+ European firms suggests that's changed: AI adoption is associated with a 4% average increase in labour productivity, with no evidence of short-term job losses. In the US, productivity hit 2.7% in 2025, nearly double the prior decade's 1.4% average. The gains are real but uneven: firms that invest in training alongside AI tools see the biggest lifts.
📚 60% of special ed teachers are using AI, and their students are benefiting
Nearly six in ten teachers of students with disabilities used AI tools to help develop individualised education plans (IEPs) during the 2024–2025 school year. For the 7+ million US students receiving special education services, AI isn't replacing the professionals; it's making thorough, high-quality planning faster and more consistent, especially in chronically under-resourced schools.
🔬 THE DEEPER DIVE
What the FDA just told us about AI in healthcare, by calling a chatbot a "breakthrough."
On March 3, the US Food and Drug Administration did something it had never done before: it handed a generative AI product its "Breakthrough Device" designation. The recipient was RecovryAI, an LLM-powered chatbot prescribed to patients for the 30 days following joint replacement surgery.
Why the 30 days after surgery matter
Joint replacement is one of the most common procedures in American medicine, roughly 1.5 million surgeries a year. But the period immediately after the operation is also when the risk of complications peaks and clinical follow-up is thinnest. Most patients go home with a pamphlet and a follow-up appointment three weeks out. RecovryAI fills that gap, available around the clock, answering questions, flagging escalation signals, and keeping patients engaged with their recovery protocol.
Why the FDA's move matters beyond this one product
The Breakthrough Device designation is reserved for tools with genuine potential to treat or diagnose serious conditions more effectively than existing options. It fast-tracks development, but it also sends a bigger signal: the FDA is developing a framework for evaluating generative AI.
Our take: PM + Risk Manager lens
From a product perspective, this is an invitation. If the FDA is engaging seriously with AI at the point of care, the design space for health AI expands significantly. The products that earn trust and reimbursement will be built with regulatory evidence in mind from day one.
From a risk perspective, "breakthrough" creates expectations. Patients trust FDA-designated products. That trust has to be earned through rigorous post-market monitoring and honest disclosure about what the product can and cannot do.
🛠 TOOL OF THE WEEK
ChatGPT's interactive visual explainers, for finally understanding it
OpenAI's newest feature lets you generate interactive visual explanations for 70+ mathematical and scientific topics. You don't just see a formula, you see it play out, with variables you can adjust in real time. For students who've never had access to a patient tutor, this genuinely narrows a gap. It's free, and particularly significant for first-generation college students and learners in under-resourced schools.
→ Try it free: chatgpt.com
💬 ONE QUESTION
When AI starts handling the administrative burden of your job, the paperwork, the scheduling, the denials, what do you think most people will actually do with the time it frees up?
Hit reply. We read every response.