Avoiding LLM feature overload

Sure, we all use an LLM every day, but when powerful new features arrive almost weekly, each promising to redefine how we work and think, why is it so hard for users to break a habit? In my experience from a user’s perspective in professional work, much of the progress feels strangely invisible.

The issue isn’t capability; it’s clarity. You kind of need to follow all the announcements closely to understand the advancements, but even then actually changing habits on basic prompting and untidy chat histories is a big issue. While most softwares have a learning curve and only a small percentage of people actually use the more advanced features, LLMs seem to magnify this issue.

Many of the most useful features — memory, custom instructions, tasks, integrations, model selection — are either buried in menus or poorly explained. Users don’t ignore them out of disinterest, but because they don’t know they exist, or don’t understand how they fit into their daily workflows.

Until AI companies focus as much on storytelling and onboarding within the apps as they do on engineering, many genuinely useful features will remain hidden in plain sight.

My top three to-dos for every basic LLM user:

1. Projects. Not just a great way to organise related chats, they actually share memory, context and don’t compact long chats, reducing hallucinations.
2. Tasks Most people I know still don’t realise you can automate simple things even on LLM free plans. Literally just ask it to do a thing at a certain time!
3. Interview yourself. When it comes to custom instructions people think it’s too much hassle to give an LLM a comprehensive doc, so ask it to interview you instead and feed back your answers as a great way to train it about you and your preferences.