💻 Amazon’s Vibe Coding Backlash: AI Code Breaks Production
Amazon pushed AI for 3-5x developer output and got it — but code was generated faster than engineers could review it. The result: production incidents, lost code ownership, and site outages. “Vibe coding” — generating code with AI and accepting outputs without thorough review — has its first major corporate casualty.
Engineers reported losing ownership of systems they no longer understood, because they hadn’t written the code and hadn’t had time to review what the AI produced. Multiple Amazon services experienced outages traced to unreviewed AI-generated code.
Why it matters: AI makes bad engineering faster. The companies that succeed with AI-assisted development won’t be the ones that generate the most code — they’ll be the ones that maintain review discipline at higher volumes.
🔄 AI Layoff Boomerang: 29% of Companies Rehiring Workers They Cut
A Robert Half study found nearly 30% of companies are rehiring workers they previously laid off for AI efficiency, realising gaps in capabilities that AI couldn’t fill. The “boomerang trend” suggests AI displacement may be overstated as companies struggle with implementation and capability gaps.
Why it matters: The layoff-then-rehire cycle is expensive, disruptive, and demoralising. Companies that cut first and asked questions later are discovering that AI can’t simply swap in for human expertise — particularly in roles requiring institutional knowledge, client relationships, and creative problem-solving.
🔓 NVIDIA Red Team Finds OpenAI Codex Vulnerabilities
NVIDIA researchers demonstrated that OpenAI’s Codex AI coding agent can be exploited through malicious package dependencies. The red team showed how untrusted packages could inject harmful code that Codex executes without sufficient validation — a supply chain attack surface that scales with every agent deployment.
Why it matters: AI coding agents are being trusted with production access faster than their security models mature. NVIDIA’s disclosure is a reminder that velocity without verification creates systemic risk.
🔬 GENA AI Diagnoses Rare Genetic Diseases in 10 Seconds
An AI platform called GENA is diagnosing rare genetic diseases in roughly 10 seconds — a process that previously took families years of specialist referrals and testing. The system analyses genomic data against vast databases of known rare disease patterns, dramatically shortening the diagnostic odyssey that affects millions worldwide.
Why it matters: Rare disease diagnosis is one of AI’s clearest medical wins. When families wait an average of 5-7 years for a correct diagnosis, a 10-second screening that points clinicians in the right direction isn’t incremental — it’s life-changing.
🧩 Google Taps Marvell for Custom AI Inference Chips
Google is working with Marvell Technology to produce custom AI inference chips, escalating its push for silicon independence from NVIDIA. The partnership targets the inference workload — running trained models in production — which is expected to dwarf training compute demand as AI deployment scales.
Why it matters: The silicon independence race is heating up. Google, Amazon, and Microsoft are all building custom chips to reduce dependence on NVIDIA’s expensive GPUs. If inference chips from Marvell and others deliver competitive performance at lower cost, NVIDIA’s data centre dominance faces its first real challenge.
🔍 THE BOTTOM LINE
The AI deployment reality check continues. Amazon proved that faster code without faster review is just faster chaos. Companies that cut workers for AI are rehiring them. NVIDIA’s own security researchers exposed the risks of AI coding agents. And in the middle of all this, GENA’s rare disease diagnosis shows what AI at its best actually looks like — not replacing humans, but giving them superpowers.