A few months ago I set up a folder structure on my computer and called it Holocron. Star Wars reference, yes, but the logic was earnest: a place where knowledge lives, organized well enough that I can hand it to an AI and have it understand what I'm working on, not a productivity system but more like a context layer for a collaborator who needs to be briefed every time we talk.

That project has become one of the more clarifying things I've done in a while.

The shift is structural.

When people talk about AI changing what it means to be a designer, the conversation collapses fast into anxiety. Will it replace us, will it commoditize craft, what happens to junior roles? I've had that conversation more times than I can count this year and I know I will continue to have it.

The shift I'm going through looks different, or so I think that's the case. In my role and what I've been doing is designing the human-facing layer of AI systems. Thinking through how people interact with agents, where AI surfaces in a workflow, what gets handed off to automation and what stays with a person. That is the work right now, and for the most part it's being done with our traditional products and systems with specific integration of AI tooling.

The work has shifted from designing for AI to designing with it and designing it, and those are different problems that will need very different approaches. I'm still working out where existing skills transfer and where I'm in new territory ... it's honestly a daily evolution.

I read a lot, follow the research, and pay attention to what's shipping and what's vaporware, but none of that is the same as having a stake in making something work.

The only way I've found to learn this is to build.

Holocron Ops (my first step to realizing full Holocron), is the operational layer running on top of Claude, and frankly has been the most useful learning environment I've found. It functions as my personal assistant for evenings and weekends, covering evening check-ins, weekly reviews, brain dumps, and personal project pickups. Building for my own use forced a level of specificity no tutorial could, am I doing something wrong... probably, but I'm learning.

What I've learned is mostly about context management and constraint design. The mechanics that make AI assistance useful rather than vaguely impressive. This comes down to knowing when to surface information and how much of it to surface, context management.

That carries into my day job well, because AI output reflects the quality of thinking you brought to it, and a lot of AI implementations fail there ... garbage in garbage out. Perhaps the organization hasn't structured its knowledge, its decisions, its operating model, and then wonders why the AI isn't helpful, but the gap was never in the AI, it's the org and the way the org is use AI.

The honest mid-year position...

Six months in, the discourse about AI holds less of my attention than the work of figuring it out. Breathless optimism and defensive skepticism are both more comfortable than building something and watching it break.

The designers and leaders who come through this period well will be the ones willing to be bad at something new, to experiment without a clear ROI, and to build for themselves before they build for anyone else.

Where Holocron goes from here, I can't say. I'm currently exploring local models managed by a custom interface, still holding the Star Wars theme, but tasteful. Where the field of design with AI goes, I can't say either, but I'm not watching from the sidelines, and for now that's a defensible position.