EP 065 · 84 MIN DO IT TWICE? THAT'S YOUR NEXT CLAUDE SKILL
Claude Knows Kung Fu. Midjourney's Still in the Dojo with v8.1?
The short version
Episode 65 is mostly a deep dive into how Rory and Drew actually build with Claude: when a repeated task becomes a skill, why Claude Code skills are separate from regular ones, and how Rory's 'orchestrator plus roster of skills' setup beats rigid, linear pipelines like n8n. Rory also debuts an Edge Case Detector skill and a month-long Claude interview he calls his Creative Intelligence Layer, while Midjourney news breaks that V8.1 is landing next week.
Key takeaways
- 01
Rory's rule: if a task is the same process and you'll do it more than once, it should be a Claude skill — a one-off shouldn't.
- 02
Claude Code skills are a separate system from regular Claude skills — you have to build them specifically for Claude Code, not just reuse the others.
- 03
Instead of rigid, linear pipelines (Rory's knock on n8n), build an orchestrator that picks which skills to deploy per project — easier to maintain and fix.
- 04
Rory's new 'Edge Case Detector' skill maps every way an input could go wrong (multiple products, a person in frame, a grid) before he writes the system prompt.
- 05
Back up your Claude MD files and skills in multiple places (GitHub, cloud, a drive) — Rory learned this after a Claude environment crash wiped his conversations.
- 06
Give Claude a memory file per person you work with (how you know them, their role, active projects) so it can draft emails without re-explaining context.
- 07
Midjourney's next release, considered internally as 'V8 beta,' is being simplified to the name 8.1 and was expected the following Monday through Wednesday.
Receipts · said on the record
“If you do it more than once and it's the same process, it should be a skill. That's just like hard rule for me.”
“The matrix is so on point with everything, which is like, remember when he plugs in and he just goes, I know Kung Fu, like that's what a skill is. But that Kung Fu has to be for everyone.”
“What happens if it's on a person? What happens if it comes in a grid? … It sort of lays out all the variables for me from the beginning.”
“If everyone's Claude is gonna be really good, and everyone's gonna be operating on the same kinds of things, producing the same kind of stuff, what makes us different again? It's us.”
“There's just so many different layers to it. You still have these user preferences in Claude, but that shouldn't be this deep, dense documentation. That should be reserved for something like a Claude MD file.”
“Wake me up when the edit model comes. That's to me when we're gonna really test this and see if people start to get back in the tool for more than just generating these first images.”
Chapters
- 0:00:00 Intro and the boys are back
- 0:01:12 Why the podcast got bigger than Midjourney
- 0:03:25 Updated Midjourney V8 take
- 0:04:40 Why new personalization profiles helped
- 0:06:31 The weird Midjourney magic is back
- 0:13:24 Midjourney 8.1 might be the real V8
- 0:16:36 Why the edit model really matters
- 0:18:51 Claude skills: what should be a skill
- 0:23:58 Why Claude MD files matter so much
- 0:25:23 Building flexible AI systems, not rigid ones
- 0:38:38 An edge-case detector skill
- 0:41:55 Memory files for people and projects
- 0:43:48 The creative analogy engine
- 0:50:35 Why examples beat vague prompting
- 0:53:38 Back up your Claude brain immediately
- 0:55:30 Orchestrator and skill roster
- 1:06:04 Hard-coded flows vs flexible AI judgment
- 1:07:43 Building a creative intelligence layer
- 1:18:01 William Shatner, Babe Ruth, and better analogies
- 1:20:53 Outro and comment bait
Questions this episode answers
How does Rory decide what should become a Claude skill?
His rule is simple: if a task is the same process and you'll do it more than once, build it as a skill; if it's a one-off, don't bother, since an unused skill is just wasted setup.
Are Claude Code skills the same as regular Claude skills?
No — Drew and Rory point out Claude Code skills are a separate system, so a skill built for regular Claude chats won't automatically work in Claude Code; you have to build it specifically for Claude Code.
What is Rory's 'Edge Case Detector' skill?
It's a skill that, before Rory writes a system prompt, maps out every way an input could go wrong (a product shot with a person in it, multiple products, a grid layout), so he can plan for those cases upfront instead of discovering them through failed tests.
What did Fast Hours say about Midjourney's next release?
Drew read Midjourney's own notes describing an internal 'V8 beta' being simplified to the name 8.1, expected the following Monday through Wednesday, with V8 staying available for about two weeks afterward; Rory and Drew agree the real unlock is still the promised edit model.
What is Rory's 'Creative Intelligence Layer'?
Rory describes having Claude interview him over about a month with broad, therapist-like questions (why he thinks he's creative, his first exciting memory) to build a personal 'operating system' of thought patterns and micro-skills he can feed into his agents.
Show notes
In episode 65, Drew Brucker and Rory Flynn return for another round of the Fast Hours podcast and immediately do what they do best: poke at bleeding-edge AI tools until something breaks, gets weird, or accidentally becomes useful.
They kick things off with fresh Midjourney V8 impressions, including what changed after another week of testing, why resetting personalization profiles helped, where the weird magic still shows up, and why V8 still feels like it is waiting for its real arrival. They also dig into the rumored Midjourney V8.1 update, what it could fix, and why the real test may come down to the edit model, text rendering, consistency, and whether Midjourney can pull serious users back into the platform.
Then the episode takes a sharp turn into Claude, and honestly, this is where things get deliciously nerdy. Drew and Rory unpack how they are building with Claude in real life: when something should become a skill, how Claude MD files shape everything downstream, why auditing your own setups matters, how to think beyond rigid workflows, and what happens when your AI starts feeling less like a chatbot and more like a strange little operating system trained on your habits, instincts, and creative baggage.
They also get into edge-case detection, memory files for collaborators and clients, private repos, Codex as backup muscle, analogy engines, and a growing obsession with building a “creative intelligence layer” that can carry more of his actual thought process into AI tools.
