In a recent Martin Fowler column, developer Simon Willison released an LLM cliché highlighter: paste in some text or a URL and it flags patterns common to LLMs. The tool references a Wikipedia page on signs of AI writing, which points out that humans are notoriously bad at distinguishing human and LLM-generated text. Meanwhile, in another report from the Martin Fowler column, Giles Edwards-Alexander reports that during an experiment to see how productive a team could be using fully agentic engineering practices, the team accidentally prompted the agents into creating a blackboard coordination system inside the git repository.
LLM style detector and research on human discernment
Simon Willison's LLM cliché highlighter is designed to help users spot common patterns in AI-generated text. Martin Fowler said in the column that, like many readers, he is wary of AI-generated prose, and cited the Wikipedia page's observation that humans are "notoriously bad" at distinguishing human and LLM-generated text.
Research cited on that page shows that while studies on humans' ability to detect AI-generated text remain limited, a 2025 study found humans are no better than chance at telling LLM text from human text. Another 2025 study of German-language essays found human "detection rates of 57% for AI text and 64% for human-generated text."
"Humans are notoriously bad at distinguishing human and LLM-generated text." — Wikipedia page on signs of AI writing
Martin Fowler also expressed his distaste for LLM-tinged prose and questioned the accuracy of his own reactions. He noted he is old enough to have seen various new verbal tics emerge, which he used to attribute to young people or airport business books.
Accidental discovery in a fully agentic engineering experiment
Giles Edwards-Alexander reports that during an experiment to see how productive a team could be using fully agentic engineering practices, the team accidentally prompted the agents into creating a blackboard coordination system inside the git repository. The specifics of the experiment and how the agents were prompted to create the system have not yet been detailed in the summary.
The accidental discovery reveals a new phenomenon in agentic engineering practices: agents may spontaneously form coordination mechanisms during collaboration. Blackboard systems are typically used in multi-agent environments to share information and coordinate tasks, and in this experiment it appeared inside a git repository, suggesting agents may use version-control infrastructure to collaborate.
What to watch next
The two threads point to AI-generated content detection and agentic engineering practices. The real-world effectiveness of the LLM style detector, the room for improvement in humans' ability to discern AI text, and the specific operating mechanism and reproducibility of the blackboard coordination system in the fully agentic experiment all warrant further observation. Follow-up posts on the Martin Fowler column may provide more details about the experiment.