Human Stuff

Simulators let a robot practise for a hundred years overnight. They still cannot tell you when a policy is safe to run on the real cell. That call stays with people.

A blog is a public notebook of things you learned the hard way. Here is why engineers, analysts, roboticists and designers should keep one.

Threat modeling works best as a structured argument between the people building a system. A worked password-reset example, with STRIDE and the four questions.

AI agents can close most alerts faster than any SOC shift. Deciding what an alert means for the business, and owning that call, is still a human job.

Once an LLM can call tools, prompt injection stops being a chatbot quirk and becomes a path to your data. Practical controls, and where human review is mandatory.

A 0.04 mm repeatability spec does not mean your robot puts parts within 0.04 mm. Here is where the rest of the error comes from, and who decides how much is enough.

A practical process for technical posts: one idea, a real outline, a strong opening, evidence on the page, honest sources, and several editing passes.

Skeuomorphism, flat, Material, design tokens, multiplayer tools, generative AI. What each shift kept, what it lost, and where human judgment sits now.

IMU, wheel odometry, GPS, LiDAR and cameras all lie in different ways. Sensor fusion is the craft of knowing which one is lying right now.

Tools change every few years. Rams, Norman, Nielsen and WCAG have lasted decades. Why principles endure, and why judgment stays human when AI draws the layout.