Ideas you can put to work today
Practical insights on applied AI, Claude, and agentic development from the team building with them every day.
The end of borderless AI: what the Fable 5 case reveals about the future of regulation
Until 12 June 2026, AI models operated in a relatively open space. That space closed — and the precedent rewrites the rules for the entire global industry.
Gemma 4: how MoE and QAT let a 26B-parameter model run in 15 GB of RAM
Gemma 4 redefines local AI: MoE architecture and QAT training make it possible to run an elite reasoning model on a 16 GB laptop. RAM, not parameters, is the metric that matters.
Chrome DevTools for agents 1.0: the browser stops being a black box for your AI agent
Chrome DevTools for agents reaches stable 1.0 and gives your coding agent the visibility to verify, debug and optimise code in a real browser. The bottleneck is no longer generating code — it is verifying it.
All Articles
Browse the archive by topic.
Gemma 4: how MoE and QAT let a 26B-parameter model run in 15 GB of RAM
Gemma 4 redefines local AI: MoE architecture and QAT training make it possible to run an elite reasoning model on a 16 GB laptop. RAM, not parameters, is the metric that matters.
Chrome DevTools for agents 1.0: the browser stops being a black box for your AI agent
Chrome DevTools for agents reaches stable 1.0 and gives your coding agent the visibility to verify, debug and optimise code in a real browser. The bottleneck is no longer generating code — it is verifying it.
Jailbreak or pretext? The Fable 5 case and the question nobody wants to answer
The US switched Fable 5 off over a narrow vulnerability. GPT-5.5 has a documented universal jailbreak — and remains available to everyone. An asymmetry the industry needs to debate.
Anthropic redeploys Claude Fable 5 after the US lifts export controls
Anthropic restored global access to Claude Fable 5 on 1 July 2026, after 18 days of suspension under export controls tied to a cybersecurity finding.