Let's information architect the hell out of your context window

So this has been coming up in various places, let’s call it context architecture.

1. A new client (I do some consulting)
2. In my own work on Sputnik Intelligence
3. During a book review

The challenge is basically: teams are working with agents (not just engineering teams either). Agents need context. Context gets messy fast.

It’s the kind of thing that really quickly brings out the information architect in people, if you have that background. A few thoughts:

1. Context is really useful and also gets messy fast. First you throw stuff at the agent. That works. Then the context gets messy and you need something. Let’s call it information architecture. Maybe you need a little organization. Maybe a small controlled vocab or a set up rules on how content is organized. How do we handle memory? And then before you know it you might want to find out, is all this context even useful to the work? Is it being used by agents? How do we eval it? (The final lesson in the ux.md series goes over that, and hey it’s free!)
2. Man am I excited those information architecture skills are back!
3. What we’re really doing when we’re building all this context and skills and memory and stuff is we are embedding cognition into the infrastructure. That’s kind of scary, and that’s exactly why we should give all this some extra thought, and this is also (part of) how we build AI systems that are good for humans instead of evil gremlins that suck the life out of life.

So let’s get good at this. I’m gonna label this context architecture unless someone gives me a better label.

Now before I leave, this is exactly the kind of thing the Shift UX conference (yes it’s a plug) is talking about. I feel like this is the moment, we’re finally figuring out where and how all the UX skills fit and can help shape this new AI world. Have a look at their program, it’s end of September and might be the best use of your likely tiny learning budget this year, if you have it :) Do it!

Health and happiness,
Peter

PS: replies go directly to me, would love to hear your thoughts on all this.