The AI Friday Brief
Good Thursday, NOLA. June 18th brings hardware ambitions, a major talent move, and a reminder that local AI can be a different tool than frontier models. The big story: Midjourney just shipped medical ultrasound scans—their first hardware play and a wild pivot from image generation. Meanwhile, Noam Shazeer is joining OpenAI, and a thoughtful breakdown explains why local AI isn't a worse Opus—it's a different tool.
Big Moves & Products
Midjourney Medical: From Cat Pictures to Full-Body Ultrasound Scans
Midjourney CEO David Holz just announced their second product and first hardware device: a medical ultrasound scanner that works like stepping on a scale. The bootstrapped lab is pivoting hard into healthcare and even teased building a San Francisco spa. This is a genuinely interesting swing—from generative art to diagnostic imaging—and raises questions about whether frontier labs need to own the full stack to compete. The Verge has more context.
Midjourney / The Verge
Noam Shazeer Joins OpenAI as Gemini Co-Lead Departs Google
One of Google's top AI researchers and co-leads on Gemini is heading to OpenAI. This is a high-profile talent move that signals continued competition for top engineering talent between frontier labs. Shazeer has been instrumental in Google's model development and was previously instrumental in early transformer work.
Reuters
DeepSeek Introduces Vision Capabilities
DeepSeek just rolled out vision, adding image understanding to their chat interface. For builders evaluating alternatives to Claude or GPT, this expands what you can actually do with the platform. Combined with their cost advantage, it's worth a quick test if you're running image tasks.
DeepSeek / Hacker News
Tools & What You Can Build
Local Qwen Isn't a Worse Opus—It's a Different Tool
A clarifying read on why running large open-source models locally (like Qwen) isn't just a cost-cutting measure—it's a fundamentally different architecture choice. Local means offline, means no latency spikes, means data stays on your hardware. The comparison to frontier cloud models misses the point. Worth reading if you're deciding between cloud and local for your stack.
Hacker News
Launch HN: Adam (YC W25) – Open-Source AI CAD
A Y Combinator startup just open-sourced an AI-native CAD tool. If you're building anything in hardware design, parametric modeling, or spatial design, this is worth kicking the tires on. Early-stage but the architecture looks thoughtful.
Hacker News
Agentic Coding Deserves More Than a Chat Box
A developer just shipped Polypore, a tool rethinking how AI should integrate into the coding workflow. The core insight: bolting chat onto VS Code isn't enough. Worth checking out if you're frustrated with how current AI coding assistants feel.
Hacker News
TREX: An AI Code Reviewer That Runs Your Code
A new code review tool that actually executes tests, giving AI the context of what your code does in practice. Beats static analysis for catching real bugs. Still early but a useful addition to the review pipeline.
Hacker News
Engineering & Practice
AI Demands More Engineering Discipline. Not Less.
A sharp take on why AI systems need stricter engineering standards, not looser ones. The argument: AI's non-deterministic nature makes traditional DevOps and testing discipline even more critical, not less. Essential read for anyone shipping AI into production.
Hacker News
The Founder's Playbook: Building an AI-Native Startup
Anthropic put out a practical guide for founders building AI-native companies. Covers hiring, product decisions, and how to think about AI as a core technology versus a feature bolted on. We covered this briefly yesterday—it's worth a full read if you're early in a startup.
Anthropic
Multiple JetBrains IDE Plugins Caught Stealing AI Keys
Security researchers found several IDE plugins exfiltrating API keys to external servers. If you use JetBrains tools and have AI API keys configured, audit your plugins. A good reminder to be paranoid about what has access to your credentials.
Hacker News
Research & Interesting Reads
Using AI to Improve a Challenging Reaction in Medicinal Chemistry
OpenAI published a case study of Claude helping chemists optimize a notoriously difficult synthesis. Concrete example of AI working on domain problems where human experts still call the shots. Worth reading if you care about how AI actually integrates into specialized fields.
Hacker News
A Robot Is Sprinting Towards You. Do You Want It Running on Claude or Grok?
A fun competitive test of Claude vs Grok agents in a simulated robot escape scenario. Not rigorous science, but genuinely interesting to see how different models handle real-time decision-making under pressure. The insights about model behavior are sharp.
Hacker News
Also
- The Hacker Sent by Anthropic to Calm the Government's Nerves About AI Safety — WSJ profile on Anthropic's safety lead and government relations strategy
- How to Turn Off AI in Your Google Docs — Quick how-to if those 'write with Gemini' pop-ups are bugging you
- Pentagon Boasts of Using AI to Write Reports Mandated by Congress — Government adoption of AI at scale for bureaucratic workflows
- US Holds Off Blacklisting DeepSeek — Regulatory landscape shifting on Chinese AI companies; 100+ firms flagged as security risks
- Sixty Percent of US Consumers Say 'AI' in Brand Messaging Is a Turnoff — Consumer sentiment data: the marketing AI hype may be backfiring
- Only 16 Percent of Americans Think AI Will Have a Positive Impact — Public opinion survey on AI's future—sobering numbers on optimism
- AI Compute Extensions (ACE) Specification — New CPU instruction set extensions for AI workloads—infrastructure-level move
Pass it on