EP 091 · 95 MIN WE JUST WANTED TO MAKE IMAGES
AI Gave Creatives a Crash Course in Engineering
The short version
Episode 91 is a two-host episode in which Drew Brucker and Rory Flynn describe how AI pulled creatives into engineering: GitHub worktrees, security audits and a Three.js product-shot tool Rory built to fix inconsistent lighting. They compare Claude Fable 5.2 with Opus 5.5, react to Claude Meetings and Whiteboards, show Drew's visual client handoffs, and ask whether codified skills and taste rules box AI and creatives in.
Key takeaways
- 01
Build the missing tool: Rory made a Three.js product-shot simulator with a locked shelf line, camera and lighting, because small variances across thousands of products meant more post work.
- 02
Strip the light from the reference: Rory fixed silver-foil text that read half white by flattening the input image to a lightless matte version, after about 50 attempts.
- 03
Failure knowledge doesn't offload: Rory rebuilt a failing Fable-and-Opus workflow by hand in about an hour and got a better result, because he knew what hadn't worked before.
- 04
Dictate the goal: Drew teaches teams to talk the outcome instead of typing, then have the AI ask clarifying questions to remove ambiguity before it starts.
- 05
Visualize instead of writing: Drew's on-site client handoff used visual artifacts and AI-captured screenshots, because a busy team won't read pages of prose.
- 06
Log what you reject: Drew had Claude record every design change he rejected, so new work avoids those things without locking in what he likes.
- 07
Rules can backfire: Rory's clip agent drifted once feedback became rules, and he asks whether skills confine AI the way brand guidelines confine creatives.
Receipts · said on the record
“Just wanna just wanna create some images with mid journey. And now I'm now I'm worrying now I got a scan for critical vulnerabilities happening every day.”
“you really can't master anything before something new comes out and it changes the way you work. So it's like it's like learning a new version of Photoshop every week.”
“the failure knowledge that you can't offload to the machines. Like, I know that's not gonna work because I've tried that on other projects a million times.”
“Fable five point two, too expensive. Too expensive for what it produces. It's not it's not Opus five point five when it comes to like the value per token.”
“But when you have all these constrictions, it's like working at a brand, you're like, No, you can't do that. No, we can't use those colors. No, the person can't be facing to the side. And no, the text is too big, right?”
“I think one of the most important things you can teach them is like the dictation. So you're not typing. You just get used to just word vomiting everything out. You state the goal or the outcome, right?”
Chapters
- 0:00:00 Drew and Rory return for a two-host episode
- 0:01:36 Staying tool agnostic as AI keeps changing
- 0:03:12 Creatives picking up cybersecurity work
- 0:07:03 Midjourney and making things for fun
- 0:09:29 Client work crowding out creative play
- 0:16:54 Keeping up with AI's constant learning curve
- 0:18:21 GitHub and code enter the creative workflow
- 0:22:53 Human judgment fixes a failing AI workflow
- 0:23:52 Building a consistent AI product-shot system
- 0:27:45 Fixing silver-foil text in AI product photos
- 0:30:42 Teaching AI taste without boxing it in
- 0:40:00 Fable 5.2 vs. Opus on real work
- 0:42:00 IT approval holds up connected AI workflows
- 0:46:14 Usage limits and running a solo AI business
- 0:50:57 The mental load of reading AI all day
- 0:54:46 Visualizing GitHub branches in 3D
- 0:58:19 Artifacts and visual AI handoffs for clients
- 1:06:08 Building an ad-creative second brain with Grok Bot
- 1:09:15 Finding fresh references beyond your feed
- 1:16:34 Easy AI work and the search for a challenge
- 1:20:20 One-shot prompting and dictating the goal
- 1:24:28 When AI turns feedback into permanent rules
- 1:26:45 Five Whys and reusable AI skills
- 1:30:15 Comparing AI skills before adding more
- 1:31:57 The sign-off takes a detour into Bravo
Questions this episode answers
What did Fast Hours think of Claude Fable 5.2?
Drew says Opus clears it in his experience and he saw mistakes he didn't get with Opus, though he hasn't tried it on creative 3D motion. Rory finds it fast but verbose and token-hungry, and calls it too expensive per token next to Opus 5.5.
How did Rory build a consistent product-shot engine?
Rory built a Three.js simulator of a photo shoot where the shelf line, camera distance, field of view and lighting stay fixed. He exports its renders as reference images so every product comes out with the same shadow and lighting.
How does Rory use Grok Bot?
Rory has Grok Bot scan X daily for creators who break down static ads, then files the methods it finds into a Notion second brain. He also filters by location to find Japanese creators his feed never surfaced.
Why does IT approval slow down AI workflows at companies?
Drew describes a client that must use Copilot and has a Claude Teams account but isn't approved to connect Claude to Microsoft. He says that kind of block keeps many teams at a basic level.
Why does Drew make client handoffs visual?
Drew says a busy team won't read blocks of text, so his on-site handoff used visuals, a one-page executive view and AI-captured screenshots, with text kept to what step-by-step instructions need.
Show notes
Did AI give creatives a crash course in engineering?
AI has pulled Drew Brucker and Rory Flynn into debugging code and building software tools to get their creative work done.
Drew Brucker and Rory Flynn got into AI to make images. Now GitHub branches and security audits have joined the job. Rory shows a product-shot tool and a GitHub visualizer he built to make the work usable (apparently, images come with infrastructure now), while Drew walks through visual client handoffs and the taste decisions that still need a person.
Three.js product-shot references and GitHub worktrees show the technical demands of AI creative work. Claude Fable 5.2, Claude Opus, Claude Meetings, and Claude Whiteboards bring model choices and workplace constraints into focus. Software security, visual client handoffs, Grok Bot research, and AI skill audits connect the tools to daily work.
