Claude AI Mastery 2026 - Week 4, Call 7

Full video transcript

Welcome in everybody to Gen AI University’s Claude Mastery Sprint. We’re in week four of our four-week sprint series digging into Claude and mastering it for our businesses. Tuning in with members of our community who are joining us here live for this workshop series. Any wins from this week? Any wins with Claude? We did an AI hackathon this week and built a few things with Claude on my end that I’m happy to share with y’all. But first, want to open the floor to hear from you. How are things going with Claude Code and co-work? How are you doing?

A member shared: “Yeah, I’m great. Thank you, Darby. It’s Wednesday, or Tuesday, or maybe Friday — I don’t know what up and down is anymore. But what I did is jump onto the new show horse called Clawbot. I spent two days doing the rodeo on Clawbot, and it didn’t throw me off, but I was very happy using it. A lot of safety warnings, a lot of stuff going on. Did people hear of Clawbot already? It’s spelled with a W — C-L-A-W-D-bot — and it’s already been renamed to Moxbot. I learned that a few minutes ago. It’s really a spectacle. It’s an AI that you put over all the other AIs, and you can interface with it on Telegram or WhatsApp. It also uses my Claude Code, my Claude co-work, and all the other AIs I tend to use are available within Clawbot. So I’m a little confused about what’s happening and what I should focus on learning. I spend about 80% of my time learning AI and 20% on the business itself. But maybe that’s the future — that I’m becoming more of a CEO of my own business than a worker in it. I have to have the overview and be the director of everything happening. It’s interesting, I like the development, but the show horse is also a little rough to ride. So maybe, Darby, if you know about Clawbot or Moxbot, made by Peter Steinberger, you can reflect on this development and quiet me down a bit.”

Darby responded: “I actually made a post about it inside the group — go check it out. My general sentiment is: I’m sure it’s exciting, I took a look at it, seems like it could be useful, especially if you’re more technical and understand how it applies to you. But I think there’s a lot of hype around it. I could be totally wrong, but it seems like it’s going to be a shiny distraction from what I need to be focusing on right now. I don’t see a ton of how I’d use it today with such a big impact that I’d want to spend time away from the other stuff I need to be doing. There’s more than enough tools that come out all the time, and I thought it was odd they started naming it Clawbot with a W. What you said is more interesting — the importance of approaching these tools, and why I’m still focusing on Claude. We’ll talk about how we continue going deeper with Claude in the next few months. It comes down to delegation — building out tools and systems the way you might hire a personal employee to do, or just being more effective at delegating and streamlining tasks. If you’re not a good, effective delegator, you’re going to have a hard time delegating to AI, because you won’t know what needs to get offloaded in the first place. That takes time — I’m not perfect at it either. That’s what makes a good founder and CEO: knowing what you’re good at, what you’re not, and what to focus on. I haven’t set up Clawbot. I saw 115,000 LinkedIn gurus posting the same thread about it changing the game as your personal assistant. I think a lot of it is exactly what you can already do with Claude. Yes, running it locally and having it interface that way is cool, but I know people who’ve literally built that before, so I was surprised it got this much hype. It’s kind of like the hype cycle with AI tools: something comes in, gets tons of attention, tech bros in Silicon Valley jump on it, and maybe it’s the next great thing, but I don’t see it changing my life in the next two weeks. So I’m going to see where the dust settles and keep focusing on what I’m building.”

A member asked how Darby judges what to pay attention to — intuition, or something else. Darby answered: “Some of it’s intuition, some of it’s seeing over five years how these tools flash and hype up, then either die down or stick around. It used to make sense to pay more attention when something new came out. If it’s still around in a few weeks and still incredibly useful, great. But if the first thing I see is everyone saying ‘here’s how to make $10 million with this,’ and the use cases sound like the same pitch people made about Claude Code and co-work — ‘go into a business, build a specific AI agent, set it up’ — that’s the new shiny thing. In some cases it’s helpful, because someone not in the space at all might hear about it and ask what it does. But if a business doesn’t have its fundamentals, organization, context, and systems figured out, Clawbot isn’t going to magically make their business work better. That’s not true for everybody, but that’s my take.”

Marcel, a business development and communication strategist, added a framework: “One thing we do when making choices is ask, ‘Is this in the best interest of my time, energy, or money? Is what I’m doing generating revenue?’ You’ll often realize you shouldn’t do this now. I also use a system called the four Ds: what do you need to Do, what can you Delegate and to whom, what can you Defer, and what can you Dump. Very often when someone looks at their behavior and realizes they’ve been scrolling Claude for two hours, maybe they need to stop. It’s about choice, action, outcome — what’s the reward you’re looking for.”

Another member, Chel, described feeling “gold-rush fever” — always needing to be on the cutting edge — and referenced the book One Thing, which discusses the damage of switch-tasking and the value of timed, focused work: “I say, ‘I’m going to review Darby’s Claude tape for an hour, then I have a timer, we’re done.’” Another member referenced a Diary of a CEO interview about how Steve Jobs and Elon Musk focus on one or two things per day and treat everything else as noise.

Assad then asked for a mental infographic of what to use for what: skills in a project, Claude Code, Netlify, or a connector directly from Claude versus GitHub. Darby answered: “My approach is having specific projects set up for things with context — projects are fundamental, same as ChatGPT. Skills are for stuff you’ll do repeatably that leads to a specific outcome — think of them like the old mega-prompts, but now Claude knows them automatically at a keyword. Connectors give Claude additional tools to execute on those skills. For example, if I’m doing research for direct outreach, I use a skill with a Firecrawl MCP connector to web-scrape and find people. If I want Claude to update a spreadsheet, it needs the connector to that spreadsheet, not just the skill — otherwise it can’t act on it. Claude is also going further into this: interactive tools inside Claude were just shared on our last mastermind call.”

Assad pressed further on when to use Claude Code versus GitHub versus Netlify specifically. Darby explained: “I move to Claude Code once I have the idea and the architecture in place for an app or web page — something I’m not just sharing a Claude link for. I host the code in a GitHub repository, and Netlify — same as Vercel or AWS — is just a way to host and display it so others can see and access it. I use Claude Code to take instructions I’ve built inside regular Claude and actually build the application. This past weekend at the hackathon I built three or four HTML-based landing pages, updated our website, and updated a couple of offer pages this way: write the idea and context out in Claude, open a GitHub repository for the new page, give Claude Code the whole list of instructions and the knowledge base from Claude, then have it write the code so I can plug it into Webflow or connect it to Netlify to deploy, like we did with Side Hustle.”

Assad clarified his workflow: build the first set of instructions inside a Claude project, then take it to Claude Code to build the rest, then modify the actual code from there. Darby added that with Claude co-work now available, they’re still learning how it fits into their flow, since they primarily use the desktop app and keep local files inside projects rather than co-work. Day to day, most of the value comes from having the right connectors set up (Google, Gmail, Zapier), and turning repeatable opportunities into skills so they don’t have to hunt down old recipes from a notes app.

Charles offered his own function-based framework: it doesn’t matter which LLM you use for coding. Projects have a siloing effect, so use a project for context. Claude is built for skills, but skills are portable — they’re basically markdown files you can port to ChatGPT too, as long as you have a way to point to and store them. A skill is simply saying “I want to repeat this multiple times in the future” — you can literally tell Claude to make a skill out of what you just did, and it does, then point to it in your CLAUDE.md file, the first file Claude looks at. From there: a repository (local git or GitHub) packages the project so it can be picked up elsewhere, and hosting (Vercel, Netlify, etc.) is a separate function for deploying and sharing a link, requiring no computer on the viewer’s end.

Later, Bo/Peter asked about a GitHub tool called “get done” that spawns multiple Claude Code agents to run a task list, and whether it could call an image model like Nano Banana Pro to generate images automatically. Darby hadn’t tried that specific integration but described hearing about a similar workflow from a hackathon collaborator who used the Gemini API to scan video and audio files and auto-edit family montage videos — something Claude alone can’t do (MP4 reading, audio scanning, image generation), but Gemini’s API can, and it’s possible to call that from inside Claude Code. Another member, Bruce, mentioned “Anti-gravity,” an extension that ships with Claude and Gemini 3 Pro built in and can call Nano Banana directly by simply asking, plus a newer video-editing feature integrated with ElevenLabs, though he couldn’t recall its exact name on the call.

A member then asked for guidance on a health-assessment chatbot skill she built in Claude, designed to take users through a vision-enabled assessment (including photos of their fridge), generate a report, and trigger automations through Go High Level. She’d been told to use Voiceflow (too expensive per assessment), then Go High Level, then a developer suggested Lovable, leaving her unsure of the right stack. The group — Darby, Thorston, and Charles — discussed the tradeoffs: Lovable can build a front-end mockup quickly; Claude Code can build the same thing while keeping code ownership; and no-code platforms come with terms-of-service risk since you don’t own the underlying code. Charles recommended locking a product requirements document (PRD) with Claude in plan mode (on Opus) before engaging any developer, upgrading to a higher-tier Claude plan to remove usage constraints, and staying strictly in plan mode — never letting Claude jump to execution — until the PRD is detailed enough to hand off with minimal explanation.

Darby then demonstrated a live proof of concept: publishing a Claude artifact that runs a skill, takes a business description as input, and returns a downloadable report — a fast, no-infrastructure way to test whether the member’s assessment concept would resonate before building a full app. During the live demo, a Notion-connector sync attempt failed (flagged as a real limitation to design around), and Claude was unable to reliably trigger a PDF download or send PDF attachments over Gmail via MCP — both noted as known friction points. Darby suggested that for MVP testing, removing the friction of a full funnel and just sharing an artifact link is enough to validate whether people find the report useful, and that a lighter-weight version could route captured emails into Go High Level via an MCP connector for follow-up automation.

The conversation closed with member wrap-ups: Darby shared that the recent hackathon was spent building a new company website, and confirmed the Claude Mastery Sprint calls will continue into February and March, alongside a separate, deeper AI implementation intensive group that already meets before these calls. Scheduling feedback favored earlier calls over midday. Darby also previewed a planned six-week series covering a book-writing intensive and hackathon focused primarily on non-fiction business books — covering everything from operational documentation to books written to help members get booked on stages — along with new book-writing software tools currently in development.