We all know how LLM Agents work: We type it a message and it responds with text, but is that really the end all be all? Let’s examine some creative ways to do this.

I am all about removing distractions. I have a really strict calendar where I have specific times I need to go heads down and do things like write these posts. During those times I don’t want to be notified unless it's an emergency. I also have dedicated time set aside for responding to my consulting clients, and a separate time for responding to in the community and comments or questions about my writing.

Now what if I connected my LLM agent to Discord or anywhere else I might get pinged (Not email, turned those off a long time ago) and turned off all my notifications from the apps. Then I fed the LLM the rules of when I should get notified about what and when. Rules like:

Between 9-11am any notifications that are not from a 1 on 1 client should not notify Matt but should be added to a queue to be responded to later. Between 11-12 all notifications and comments from readers of the newsletter should be aggregated for Matt to respond to. Between 1 and 3 only notifications from the client Matt is doing advisory and oversight with at that time should come through and all else should be stored for the appropriate time.

There are 2 interesting things about this approach:

First my actions are not what triggered my LLM Agent to start processing, someone else's is.

Second, the time of day and my calendar 2 important pieces of context the LLM would need to fulfil its duties.

It makes you wonder how many other applications there are for this technology.

What other actions could trigger the LLM Agents?

What other dynamically changing context could be given to the LLMs to help them decide on an action?

And with the advent of Model Context Protocol what actions, besides just responding in text, could the LLM Agents take on your behalf?

Let me know what you think.

PS: I am thinking about starting a Master Mind around this. Let me know if you would be interested. ~Cheers!