Self-Learning Systems
Every new system will eventually become a self-learning system. It will have awareness of what it’s capable of and where its boundaries lie, what value it creates for you, and what worked versus what didn’t.
This is new.
We’re already used to tools helping us with planning and doing things in varying degrees. But reflection, not just evaluation and situational awareness, used to be uniquely human. LLMs are helping us sort things out, identify patterns, and make qualified guesses. Yet, the last mile remains ours: the feelings we feel and the feelings we strive to evoke.
That boundary matters. A system can optimize efficiency, but it cannot decide what’s worth optimizing for. It can surface patterns, but it cannot tell you whether those patterns align with what you actually care about. This is the in between space where human judgment remains irreplaceable and intentional design matters.
Don’t write off apps just yet
Where the new and old meet. There’s good reason for being hysterical about how easily existing functionality can be one-shotted or rebuilt with now minimal effort. At the same time I’m even more interested in genuinely new experiences we’re now able to create.
Apps represent unique spaces that provide context, store data with a purpose, and make intuitive use of input modalities. It’s a powerful package. Adding self-learning capabilities lets apps reimagine themselves: adapt to the people using them, become easier to use over time, broaden their audience, and create value that aligns with your actual aspirations rather than generic defaults.
Getting Started
The first step is deceptively simple: tell the LLM not just to do things, but to document what it learns along the way in a learnings.md file. At first, it might just capture barely-useful patterns like “user prefers detailed summaries” or “this workflow works better in the morning”. Small observations able to compound over time. Observations that don’t remove your need to reflect but that will trigger just that.
You’ll be delighted to experience how actions become more efficient and relatable when built on practical, shared insight.