I hereby give Claude my full, unequivocal, unwavering, signed in blood approval to do the thing I asked it to do without the need to repeat its' permission request on every turn on the off chance I changed my mind in the past 90 seconds.
favorite new work hobby:
1. "Read [site or collection of sites or repos] and reverse engineer everything about how they run [marketing, etc.]"
2. Read the report.
3. Ask lots of questions about its findings.
4. Figure out how we can bake the best ideas into my agents' jobs.
5. Build the custom skills.
6. Deploy using @bot and Claude.
Taste is our job.
Replying to @CasJam
the challenge with UX is that models still lack taste. Backend feature code it can ship all day long with a lot less supervision. Interested to hear how you might tackle that.
My next youtube video will show my UI/UX design process for this new app I'm working on.
I already know it won't drive many views. No buzz-wordy title. No shocked face thumbnail. No "game changer" promise.
Don't care.
Design is too important (imo). It's finally possible to use your brain (!) plus AI in UI/UX craft without spitting out vibe-coded slop. Worth sharing.
New approach on this product:
1. Weeks of planning & research
2. Wireframes → mockups with Paper MCP
3. Turn mocks into functional spec docs
4. Build a phased roadmap
5. Build one phase at a time. I review and course correct on each milestone.
Previous builds skipped the design planning phase. This product—given how interconnected it is—really benefits from it.
I'm a lot more productive when my prompts don't dictate what to do, but rather express where I'm seeing friction or a problem.
Then ask for suggestions on what the fix or improvement might look like. Then back-and-forth until we agree.
THEN build.
I never use 3rd party agent skills. But I do:
- Spend many hours crafting my own.
- Have agents analyze others' public skills & sites to find ideas we might use.
- Iterate constantly until the workflow runs the way I want.
3 goals for this dashboard:
1. Ideas list + Active projects front and center.
2. Keyboard navigation for everything.
3. Agent sidebar always one key away.
Since going back to all-in on Claude (opus 5.5 full time), I haven't hit a limit once. I'm a heavy multi-tasker.
Step 5 is key. Just because models can now build a lot in one shot doesn't mean they should.
Plan for stops so you can review and refine in stages before it goes off the rails.
New approach on this product:
1. Weeks of planning & research
2. Wireframes → mockups with Paper MCP
3. Turn mocks into functional spec docs
4. Build a phased roadmap
5. Build one phase at a time. I review and course correct on each milestone.
Previous builds skipped the design planning phase. This product—given how interconnected it is—really benefits from it.
New approach on this product:
1. Weeks of planning & research
2. Wireframes → mockups with Paper MCP
3. Turn mocks into functional spec docs
4. Build a phased roadmap
5. Build one phase at a time. I review and course correct on each milestone.
Previous builds skipped the design planning phase. This product—given how interconnected it is—really benefits from it.
Designing custom keyboard mappings.
Instead of just telling you there's a conflict, I'm giving you the option to set (or reset) both right now.
One of my more technical tool designs. Coming together so fast. Amazing how much more ambitious we can be with todays models, harnesses and MCPs!
Tiny UI detail in Claude Desktop: It sometimes just replies with an emoji and just does it (in this case, does it in @paper) 👍
Most apps do more/less the same thing. Users stick with the one who's interface fits their brain.
So:
1. There's always room for a different take in an established category.
2. UI/UX taste and product vision is still a valuable human skill.
How I'm using @bot as a team running growth tasks on my products.
youtu.be/RM8ZwhoJdeg
Pretty magical to watch Claude build wireframes in @paper in realtime.





















