Posts

Prompt injection: the AI agent vulnerability no patch can close

When an assistant reads emails, web pages, and documents, any text can turn into a command. Why prompt injection has become the main security problem of systems built on language models, and how to limit the damage.

Windows 10, last call: what to do before extended updates end

Mainstream support ended in October 2025, and the extended security updates program for home users closes in mid-October 2026. The real options for anyone still on a Windows 10 PC, from moving to Windows 11 to the alternatives.

Smartphones that last longer: what the EU ecodesign rules change

Guaranteed updates for years, available spare parts, tougher batteries, and a label that grades repairability. Just over a year after the EU rules took effect, here's what to look at before buying a phone.

Smartphones via satellite: when the signal comes from space

Messages, location, and eventually data, with no ground antenna for miles around. Direct links between satellites and ordinary phones are leaving the experimental phase. What already works, what doesn't, and why it matters even for city dwellers.

Digital coworkers: what changes when an AI agent has an account like yours

No longer a tool you open when you need it, but an identity that works autonomously inside the organization, with a calendar, an inbox, and access to internal systems. What it actually means to bring an AI agent onto a team.

Assistants that remember: what changes when AI stops starting from zero

After years of conversations that vanished with every new session, AI assistants are starting to carry around a memory of who you are and what you've already said. It's convenient — but it relocates a question that had so far been easy to postpone.

World models: the AI that tries to understand the world instead of just describing it

A language model predicts the next word. A world model tries to predict what happens next in a real scene. It's a precise technical difference, and it's why robotics is looking elsewhere than chatbots.

Synthetic data: AI increasingly trained on data it generated itself

The web's supply of human-written text no longer satisfies the biggest models' appetite for data. The industry's answer is to generate the missing data with AI itself — a convenient fix that carries a risk rarely discussed.

Digital sovereignty: what moving your data to Europe actually means

European cloud, models trained in-house, data residency requirements. Behind a word that has become a slogan sit concrete technical choices — and trade-offs worth understanding before adopting them.

Coding agents: when the AI stops suggesting and starts writing

Developer assistants have gone from completing a single line to opening pull requests on their own. Here's what actually changes in day-to-day development work, and which habits become non-negotiable.

Custom AI chips: why Big Tech is designing its own processors

Google, Amazon, Microsoft, and Meta are investing billions to build proprietary AI chips and cut their dependence on Nvidia. Here's why it makes sense, and just how hard it really is.

Agentic browsers: the AI that browses the web for you

The new generation of AI-powered browsers doesn't just answer questions. It clicks, fills out forms, and completes purchases on your behalf. Here's how they actually work, and where they still trip up.