Context is king and its loyal subjects

When we give gen-AI a task, we’re told to give it context. Context is king! Context will make Claude work for you. Context will solve all the world’s problems. So we give it everything we’ve got, and it hands us back something that could belong to literally any app on the internet. Turns out the context we’re giving it is actually making things worse.

What happens in the talk

Putting context together in a structured, well-designed way is the first and most important step to getting any real value from AI. If the context we give is too much, too broad, or formatted in a way the model can’t really use, that confuses the machine, and a confused machine gives you unexpected, inconsistent or very generic output. But if we create structured, strategic context bundles, we can move away from writing one-off prompts and toward building something more like an information infrastructure, based on the professional knowledge people already have in their heads.

The talk is built on stories and before-and-after examples from Michal’s own work. She asked AI to write the same onboarding screens for a fintech app again and again, each time with more context, to show how that context enriches the result.

What you will leave with

• Why AI keeps missing the nuances that experienced people resolve in seconds
• What goes into a context design package, and how to structure it so the model can actually use it
• How to build and maintain reusable packages, with good ROI, so a whole team gets more consistent, higher quality work

Who it’s for

Designers, content designers, UX writers, PMs, and anyone on a product team who already uses AI every day and is a little tired of getting the same bland answer back.