AI models produce uninspired designs - even Opus 5.5, despite having made huge strides, still creates relatively bland designs. It can also struggle to explore truly novel ideas when writing, coding, brainstorming, etc. Over the past few weeks I’ve been using a new /exploration skill that’s made an enormous difference with a relatively simple set of changes. The skill is, first and foremost, for doing wide, creative explorations... something I find these models still struggle with. It has three critical components: (1) a random string generation script used to generate seeds for each exploration. It uses each seed string to draw inspiration for the task at hand. (a subset of the string might influence the color palette, for example) (2) a set of reference docs that tell the model how to read the seed based on the type of task (design work, writing, marketing campaign generation, etc) (3) instructions for leveraging the @fal mcp when exploring visual/audio concepts You can see a set of example designs below generated with and without the /exploration skill using this single prompt: "create 4 landing landing page concepts for a personal AI assistant called Nova" While you can't fully appreciate it from the static images (check the link), several examples with the /exploration skill include videos and realistic images without me ever prompting the model to include them. And the model knowing it has access to videos and any style of image enables it to explore a wider variety of directions. Overall, they end up being far more interesting takes. Sharing the skill here in case it's helpful for you all! Give it a spin! 👇

And GPT 6.1! "Smarter than Astra in some ways" and cost 20%!

Say it with me: OKRs are not strategy. They're a measuring stick for your strategy... and it's hard to measure against something that only exists in people's heads. I think your strategy should be a few pages of narrative instead. Full thoughts below