Good Morning and Happy Tuesday. Welcome to issue 58 of The Tech Stop.

Exploring the pitfalls of AI reliance, tips for secure AI usage, the impact of government restrictions, and the future of AI research.

1. Ford abandoned AI-only quality inspections after defects slipped through, achieving improved quality by combining experienced engineers with AI. It highlights the risk of replacing inspection, quality, or safety staff with AI alone, reinforced by data showing many firms later rehire employees cut because of AI. Meanwhile, major companies are imposing AI spending controls after unexpected costs, underscoring the need for governance over AI tool usage. In contrast, Nestlé successfully used AI to automate interview scheduling, demonstrating AI's strongest value in workflow automation rather than replacing critical human judgement. More, more and more.

   
   POV: Adopt an AI-first augmentation policy, requiring 12-month trials before AI-driven headcount reductions. Audit AI spending and governance, and pilot AI-assisted recruitment scheduling with clear ROI and productivity metrics before wider deployment.
   

2. 41% of employees receive no AI tools or guidance from their employer, driving 76% to self-source unapproved tools with no visibility on what data is being shared externally. This creates data security, liability, and regulatory exposure. More.
   
   POV: Commission a joint HR and IT audit to surface which AI tools employees are currently self-sourcing and for what tasks. Use findings to deliver a sanctioned AI toolkit covering the most common workflows, and an acceptable use policy specifying which data categories must never enter third-party models.
   

3. US Government restrictions on frontier AI models have already disrupted enterprise access, creating a new operational risk for businesses reliant on specific AI systems. At the same time, Microsoft, Amazon, OpenAI, and Anthropic are investing heavily in embedding engineering teams within organisations to help deploy AI into production rather than pilots. This increases vendor dependency, operational resilience concerns, and intellectual property risks that require careful governance and commercial planning. More, more and more.
   
   POV: Map critical AI dependencies, require fallback providers, and review contracts for resilience. Evaluate embedded engineering offerings from major vendors, while establishing standard legal positions on data ownership, IP, and exit rights.
   

4. AI research is shifting from language models toward "world models", systems that understand physical environments, geometry, and physics with billions in investment. They are being used to generate synthetic training data for industrial robots and autonomous systems, directly addressing the data shortage that limits physical AI deployment on factory floors and construction sites. More.

   
   POV: Assess whether planned robotics or autonomous equipment investments could benefit from synthetic data generation tools, particularly for safety-critical edge cases that physical systems rarely encounter in the real world.
   
   Something to keep an eye on: A flywheel-based energy storage platform designed for power-intensive applications can be deployed to meet the power demands of AI data centres. 
   
   And that was the world of tech last week - join us same time, same place next week 🥳