What if your eye could detect schizophrenia?🌿

The FDA is redefining what "breakthrough" means for AI medical devices, shifting the standard from "faster than a doctor" to "does something a doctor simply cannot do." Think: detecting multiple cancers from a single image, or predicting a patient's probability of dying from heart failure within five years. This is regulatory maturation happening in real time. Rather than approving AI tools that mirror existing care, regulators are pushing for technologies that genuinely expand the frontier of what is medically possible. For anyone anxious about AI in medicine, this is a signal that the oversight infrastructure is keeping pace.

A landmark Ember report finds that AI-powered grid management across Southeast Asia could deliver $67 billion USD in cost savings and eliminate 400 million tons of CO2 emissions between 2026 and 2035. Solar and wind already cover 5% of ASEAN electricity, up from 2.3% in 2020. AI forecasting and predictive maintenance could push that to 42–47% renewable by 2045. Indonesia, Vietnam, Thailand, Malaysia, and the Philippines all score above the global average on AI readiness. The infrastructure is there. What is left is policy and scale.

AI to unlock the next wave of renewable integration in ASEAN | Ember

New World Economic Forum data offers a rare piece of genuinely encouraging labor news: candidates who list AI skills on their resumes are 8–15% more likely to receive an interview invitation, across fields as varied as graphic design, office admin, and software development. The biggest relative gains went to older applicants and those without advanced degrees, for whom a recognized AI certificate made the most difference. In South Asia, AI-related roles pay roughly 30% more than comparable white-collar positions. Upskilling is emerging as a real ladder, not just for those already near the top.

The First AI-Designed Drug Just Proved It Works. Fifteen More Are Coming.

Here is the number that stopped us in our tracks: patients taking rentosertib gained an average of 98.4 mL of lung capacity. The placebo group lost 62.3 mL. That is a 160 mL gap, in idiopathic pulmonary fibrosis (IPF), a disease where losing lung function is the expected story and where most treatments only slow the decline. This gap is not incremental. It is a signal.

What makes rentosertib different from every drug in every other trial right now is how it was found. Insilico Medicine's generative AI identified both the disease target and the molecular compound, without human intuition in the discovery loop. It is the first drug to complete that end-to-end AI journey and show real results in humans. Results published in Nature Medicine.

This matters because drug development is broken in almost every way that counts. It takes an average of 10–15 years and over $2 billion to bring a drug to market, with a failure rate exceeding 90% in clinical trials. AI-first platforms promise to compress the discovery phase from years to months, and to find candidates that human researchers might never have considered. For years, that promise lived mostly in press releases. Rentosertib is the proof-of-concept that the field has been waiting for.

The pipeline behind it is real: as of early 2026, 173 AI-discovered drug programs are in clinical development globally. Between 15 and 20 are expected to enter pivotal Phase III trials this year alone.

Our take, through a PM and Risk lens:

From a product perspective, this is a platform story, not a molecule story. The value is not rentosertib specifically, it is evidence that the underlying AI discovery engine works. Every successful trial makes the next candidate faster and cheaper to find. What starts as one impressive result becomes a fundamentally different model for pharmaceutical R&D.

From a risk perspective, the question we are watching is regulatory novelty. Agencies like the FDA and EMA are now being asked to evaluate drugs where the answer to "how did you find this?" is "the AI identified it." That is not a reason to slow down, but it does require new frameworks for auditing AI-generated discovery pipelines to ensure reproducibility and safety at scale. The governance infrastructure needs to grow as fast as the science.

Looking ahead, we expect Phase III data for 2–3 AI-designed drugs by the end of 2027. If even one achieves full regulatory approval, it reshapes the entire pharmaceutical R&D model, shifting major investment from large internal discovery labs toward AI-first biotech partnerships. The companies building these platforms today are laying the infrastructure for the medicine of the next decade.

SpeciesNet is Google's open-source AI model trained on 65 million wildlife images that can identify nearly 2,500 species of mammals, birds, and reptiles from camera-trap photos. It detects animals in 99.4% of images and achieves 94.5% species-level accuracy. One researcher at Wake Forest University used it to process 11 million backlogged photos in days, work that would have taken human researchers years. It is freely available on GitHub under the Apache 2.0 license. For conservation teams working without large budgets, it is a transformative tool.

→ Read more: How our open-source AI model SpeciesNet is helping to promote wildlife conservation

If AI can now design drugs, screen for mental illness through your retina, and optimize energy grids across entire regions, what is the most important problem in your world that you think AI still has not touched?

Hit reply. We read every response.