@Rafal Glaz Fair point on sources. I went and had a look; - Wistia's 2025 State of Video Report (14 million videos, 1,300 professionals surveyed) found software and tech videos perform best at 3-30 minutes. Mike's 5-min sits in that range - https://wistia.com/learn/marketing/video-marketing-statistics - Sight AI's VSL guide says screen recording is best for software and digital products ("show, don't tell"), and talking head is best for coaches and personal brands - https://www.trysight.ai/blog/video-sales-letter - Contesimal's VSL guide notes that some of the highest-grossing VSLs of all time are completely faceless - https://contesimal.ai/blog/video-sales-letters/ Worth highlighting though, working in B2B mental health training, I refer my clients to a product in the same space (cpdpassport.co.uk, where I have an affiliate relationship). So Mike's world is one I'm in. That probably explains why I was happy watching his video regardless whether he was on screen or not, I was watching as someone already curious about training and compliance software, not as a cold stranger. I hadn't thought about that until later on. @Mike Bryan, if you need feedback that maybe predicts conversion, run it past people who match your buyer (HR managers, compliance officers, safety leads in regulated industries). I'm sure they'll be more than happy to give you feedback on which format works better or maybe create a few different versions to use depending on how and who you're using it with. I have 3 clients who are great at giving me feedback. The cold-vs-warm question is a good point. For cold ads, more of Mike on screen probably helps. For warmer traffic (referral, someone problem-aware), the screen-recording version holds up well.

@Mike Bryan Good luck! Keep us updated on how you get on.

@Rafal Glaz yes, exactly, I have 11 active n8n automations running, which is a separate server humming away 24/7. That's why they don't show up in my Claude Code usage. Claude Code is where I build and edit the workflows, n8n is what runs them once they're live. The only AI cost sitting inside the automations is one workflow that uses Straico to summarise emails before they land in my CRM, and that comes off a separate Straico balance I already paid for. So the $100 really is just covering my hands-on building, drafting and briefings. The automations themselves run on my Hetzner server for about €5.50 a month.

@Rafal Glaz almost certainly yes, it would max it out fast. But more importantly, it'd be the wrong tool. 10 of my 11 automations don't need AI at all, they're just moving data between services on a trigger. Running them through Claude would mean paying for thinking on tasks that need none. n8n is the right tool for plumbing, Claude Code is the right tool for the building and the thinking work.

Good that the agents caught it @Nesil Ozer. What were they fetching when it happened? Same kind of site or a mix? And are you keeping a log of the attempts? Useful to spot patterns if it keeps happening. Not something I've run into yet so curious what you saw.

@Nesil Ozer Oh, I see! 😅

@Rafal Glaz Very, it's crazy eh. We usually buy Crucial or Corsair, but Crucial decided to exit the consumer‑memory market and is no longer selling its modules directly. I'm glad we got ours when we did.

@Rafal Glaz As far as I know they're not going to come down anytime soon either. What's required for AI is a long game.

Hey @Mikke Papes, you're not lost, this is genuinely confusing. Here's how I'd describe the three after using them all: Claude.ai in the browser is the everyday chat. Drafting, thinking out loud, research. Skills and Projects live in your account, so they're there whichever browser you sign into. Claude Desktop looks like Claude.ai in an app window, it's Claude working alongside you, doing tasks on its own across your local files and apps. You set a goal, it gets on with it. Available on paid plans. Claude Code is the one that surprised me. I thought it was for developers but I use it for everything now, emails, content, CRM updates, the lot. It runs in a terminal or inside an IDE like VS Code, and the difference is it can read and write files on your computer. So if you keep your work in a folder, it can actually do the work, not just talk about it. Cowork is essentially the same idea aimed at non-developers. The reason skills don't cross over between Claude.ai and Claude Code: Claude.ai skills live in your Anthropic account in the cloud. Claude Code skills are just folders of text files on your computer. Two different homes for the same idea. Your GitHub instinct is sound for Claude Code skills, that's how a lot of people sync them. I do something simpler, I keep my whole working folder in Google Drive so the same skills follow me from Mac to Windows automatically. One thing I didn't realise straight away: for this to actually work you need to right-click the folder in Drive and set it to "Available offline" on each machine. That means it physically lives on your hard drive and syncs in the background, so you do need enough disk space for it. Once that's done, whichever machine I open, Claude Code sees the same files. Doesn't fix the Claude.ai side, those skills have to be set up in the web UI separately. The way I'd put it: Claude.ai is for talking to Claude, Claude Code (or Cowork if you're not in an IDE) is for working with Claude on your actual files. Once that clicked, the rest got simpler. I'm sure there are things you can't do in Cowork that you can in Claude Code using in IDE, so that is by far the best and most powerful option to learn once you've got a handle on the basics.