When humans need information, we don't load everything we know into our heads at once.
We ask questions.
We look things up.
We pull in details when they become relevant.
AI systems should probably work the same way.
Recently I updated Empirical's CLI documentation system.
Before, I could have dumped the entire command reference into every agent session and called it a day.
Instead, the installer adds a tiny instruction:
empirical doc
empirical doc <topic>
empirical doctor
That's it.
The agent doesn't get the entire manual.
It gets a pointer to the manual.
When it needs help with memory commands, it runs:
empirical doc memory
When it needs installation help:
empirical doc install
When it needs to discover what's available:
empirical doc
The detailed documentation is loaded only when needed.
Less context. Better timing.
What's interesting is that this is becoming a pattern throughout Empirical.
The CLI uses on-demand documentation.
Memory retrieval works the same way.
Conversation context works the same way.
Instead of shoving everything into the prompt and hoping the model finds what matters, Empirical tries to surface only the information relevant to the current task.
I've started thinking of this as progressive disclosure for AI.
Not bigger context.
Better context.
The future may not belong to systems that remember everything. It may belong to systems that know what not to load until it's actually needed.
This idea has become one of the guiding principles behind Empirical.
We're exploring what happens when AI systems retrieve information as needed instead of carrying everything around all the time.
If that sounds interesting, I'd love for you to take a look at Empirical and share your feedback:
Top comments (1)
"I hit my 5h window. FML" HAH this is me. I literally just did that and walked away from my desk.
😂😂