Insights & perspectives

The new generation of language models is actively trained away from agreement. Anthropic's Claude Fable 5, relaunched in July 2026, is the clearest example so far:

The word sycophancy has circulated in AI research for several years. It appears in papers, in model evaluations and in technical documentation from the major AI companies, and over the past year it has stepped into the broader conversation about how language models affect us. The word sounds specialised, like a property only researchers can measure and only engineers can fix.

I have helped people start using AI. Over the years I have carried the knowledge of language models forward, shown what they can do, opened the door, made them useful in everyday life. Part of what I set in motion is now unfolding into something else.

Ten markers, six mechanisms, seven tips for writing like a human again A few years ago I received an email that was clearly written by an LLM. It made me pause and consider what kind of communication I should respond with. If the sender did not take the time to write it themselves, what obligation did I have to spend my own time on a reply? I dropped it into my own model and sent something back. This is what much of our communication looks like today. It is becoming the norm, but is it the norm we actually want?

Cut-and-paste began as a physical editing technique. Writers and editors literally cut text with scissors and rearranged it with glue. This required reading, understanding, and deliberate reorganization of ideas.

Language builds the framework of thought. Every structure we use to speak also becomes a structure we use to think. Across languages, these structures differ in grammar, metaphor and category, creating distinct ways of perceiving and relating to the world.