This week's edition covers Google's open-source medical AI models, a developer's $125 monthly AI stack delivering 10x speed gains, breakthrough diagnostic AI agents, neurofeedback training for focus, and practical camera-based problem solving
🏥 Google MedGemma Puts Medical AI in Developers' Hands
What it is: Google's MedGemma is an open-source collection of AI models designed specifically for medical applications, available in two variants—a four billion parameter multimodal model that processes both medical images and text, and a 27 billion parameter text-only model for complex medical reasoning.
What's new: Google released MedGemma at I/O 2025, democratizing access to medical AI by providing pre-trained models built on the Gemma 3 architecture. The four billion parameter multimodal model can analyze chest X-rays, dermatology photos, histopathology slides, and ophthalmologic images, while the 27 billion parameter text model excels at clinical summarization, patient triage, and medical question answering. Developers can adapt these foundation models through prompt engineering, fine-tuning, or integration with other tools like web search and FHIR (Fast Healthcare Interoperability Resources) systems.
Why it matters: Instead of building medical AI from scratch, developers now have a proven foundation to customize for specific healthcare needs. However, early testing reveals important limitations—one clinician found the model missed clear tuberculosis signs in a chest X-ray, reporting it as normal. This highlights why these models require careful validation and refinement before real-world deployment. For individual developers and small teams, MedGemma offers an accessible entry point into healthcare AI that was previously only available to large organizations with extensive resources.
🛠️ From Weeks to Hours: One Developer's $125 Monthly AI Stack Delivers 10x Speed Gains
What it is: A detailed breakdown of one practitioner's complete AI development workflow, from foundational models (Claude, OpenAI, Grok) to coding tools (Replit, Cursor) and workflow automation, showing exactly how $125 monthly transforms development timelines.
What's working: The author documents moving from idea to functional prototype in hours rather than weeks using a specific two-tool progression: starting with Replit for rapid prototyping and AI-assisted development, then transitioning to Cursor (a VS Code fork with superior AI integration) for production-quality code. Key insight: using different tools for different phases rather than trying to force one solution through the entire development cycle. The workflow includes creating dual PRDs (product requirement documents)—one detailed for planning, one condensed for AI agents—leveraging multiple AI models for their specific strengths, and maintaining a clear handoff process between prototyping and production environments.
Why it matters: This isn't theoretical—it's a working system with documented costs and measurable time savings. The workflow demonstrates how to practically combine multiple AI tools without subscription overlap waste, when to transition between different development environments, and how to structure projects for AI collaboration. Most valuable for practitioners: the specific tool combinations and handoff points that actually work in practice, plus the honest assessment of what each tool excels at and where it falls short.
🏥 FDA Accelerates Drug Reviews with Agency-Wide AI Deployment
What it is: The U.S. Food and Drug Administration is the federal agency responsible for approving new medications, medical devices, and treatments before they reach patients. Their review process traditionally involves extensive manual analysis of scientific data and documentation that can take months or years.
What's changing: Following a successful pilot program, the FDA announced an aggressive timeline to deploy AI tools across all agency centers by June 30, 2025. The generative AI system reduces tasks that previously took scientists three days to complete in just minutes. One FDA deputy director reported being able to perform complex scientific review tasks in minutes rather than days, calling it "game-changer technology."
Why it matters: This marks the largest AI implementation in federal regulatory history and directly impacts how quickly new treatments reach patients. For individuals and families waiting for breakthrough therapies, faster FDA reviews could mean accessing life-saving treatments months or years sooner. The approach also demonstrates how AI can augment human expertise rather than replace it—FDA scientists spend less time on repetitive analysis and more time on critical decision-making that requires human judgment.
More AI everywhere doesn’t mean better results. Small teams of AI experts beat mass licenses. Attackers now use AI to run cybercrime at machine speed. Leaders can’t keep up. Tools still forget us. The winners? Focused expertise and hybrid AI, not AI everywhere.
You've built something and you need to know if it works. So you do what's sensible—you ask an LLM to grade it. Factual accuracy, code quality, agent outputs. The machine judges the machine, and you get a number you can act on. Except that number
Imagine you have a bunch of teams, some with AI, some without, and some where everyone gets their own AI. Researchers ran a big experiment with over 400 people to see what actually happens when you mix and match humans and AI in different ways.
Here’s what they found:
