✴︎ Doing hard yakka

OK Claude you want me to take a break I'll take a break...

most of my communication with Claude Code 😅

A week ago I was a complete noob.

Today I'm running 3 terminal sessions with Claude Code, managing git versions, iterating on different designs simultaneously.

The learning curve is wild.

Because most designers I know are using AI tools, but nobody's tracking consumption or encouraging us to use more.

Is it on us to prove the ROI first, or should companies be driving adoption?

Jensen Huang just said if your developer isn't consuming $250k worth of tokens on a $500k salary, something's wrong with them.

But here's the real debate: should that come from the developer pushing to use more AI, or should the company be mandating it?

I love Oscar season memes

Would an always-on display for token consumption actually solve this or am I the only one annoyed by this?

Running into this constantly:

I'm deep in a Claude session (app or terminal), then have to stop, switch to settings, check my session limit and monthly usage, then jump back.

Breaks my flow every time.

Testing....

This feels like a real product opportunity. A design challenge I want to dig into.

Tomorrow I'll go deeper into the research.

Building this publicly to see if people actually want it. If you've worked on similar problems, let me know what you learned.

One thing that keeps coming up: personalization works, but timing kills it.

Introduce it too early and you create friction instead of value.

Also learning that calculating CPW is way harder than it looks. There's real complexity here.

Found something interesting in the research:

71% of consumers want GenAI integrated into shopping experiences. 25% already used GenAI shopping tools in 2025.

People are ready for smarter shopping tools. The question is how to build them right.

Spent today researching how CPW is being used in different brands.

Dug through old Reddit threads, blog posts, existing implementations.

The implementations are all over the place. Some work, most feel bolted on.

As a designer, I immediately saw the gap: most e-commerce doesn't answer basic questions. What am I buying? What's the real cost? How long will it last?

Feels like a problem worth solving.

Starting a new project today. Building in public.

My partner owns an eyewear brand. She introduced me to Cost Per Wear (CPW), a concept that calculates the real value of what you buy.

Or maybe they are using The Force?

Am I the only one who's experiencing severe anxiety from the last seconds of the Figma tutorial on Slots? Couldn't you make them hold the button?

Or are they using The Force?

Going to clean this up and share it once it's stable.

Who wants to test it and tell me if I'm overengineering this whole thing?

Could use some outside perspective before I go deeper.

Stuck on this question.

Is the approach overcomplicated? Should I simplify what it gathers?

Or is more context actually better and the drift is a separate problem I need to solve differently?

Honestly can't tell from inside the build.

Problem: it still drifts after multiple iterations.

Better than before, but not solved.

I think I might be pulling in too much context. Interview, gather resources, listen, synthesize. Maybe that's creating something too complicated for the model to actually use.

Not sure yet.

Not perfect but it's working better than what I was doing before.

Early results are interesting.

The model is staying more consistent. Instead of generic style rules, it learned my actual decision patterns. So it remembers why I made choices, not just what.

Component generation feels more intentional too. Less generic button soup.

Not a lot of the big ones will stay relevant. New ones will grow.
Internally the working force will experience a shock (perhaps even a long winter) where Generalists and professionals with the will to change and adapt will thrive but also professions will consolidate.

Quick update: got the interview flow working.

It asks about your design approach, catches resources you mention during conversation, then merges everything into one context file.

Now testing if Claude actually uses it better than the generic context I was feeding it before.

Do you use a skill for that or just keep it clean yourself?

SaaS is not dying, it's evolving.

Honestly not sure if this will work the way I'm thinking.

But the logic makes sense. Pull context from where you already work. Let it ask questions to fill gaps. Output something Claude Code can actually use.

Worst case, I learn what doesn't work and try something else.