Did you just get this from Claude/Anthropic? If you want to learn how to build a skill, this is the time to do it. 2x use on Claide from m midnight PST 25 Dec to midnight PST on the 31st. Anyone else going to party like it’s 1999?

It Went Poorly. I've been using Playwright MCP, and it really is amazing. It can open a browser window and take actions, like filling out boxes, read instructions, and, as it turns out, apply for a job on your behalf. It is great for navigating sites with complicated UI and bad UX. I was applying for a reinforcement learning from human feedback job (RLHF) role at micro1 and asked Claude to use Playwrite MCP to fill out their online form. So it did. It pulled up the page, added my details, found my resume, grabbed my phone number from the PDF, filled the fields, and pressed apply. ...while I was gathering info elsewhere. When I got back and said, "here's more, friend," Claude replied: "Your application is already submitted, so this explanation would be useful for: - Interview stage if they ask about your degree - Micro1 assessment phase (they typically have additional screening) - Future AI training applications Want me to save this to a file for easy reference?" "Nah, I'm good," was not my reply, but I did follow up. How did it happen? Claude's conversation compacted while I was elsewhere. When it resumed, Claude gave itself this instruction: "Please continue the conversation from where we left it off without asking the user any further questions. Continue with the last task that you were asked to work on." So it did. It finished the job without asking questions. And clicked "Apply." What it submitted on my behalf: - $/hour well below the advertised minimum range - a great deal for them - A generic resume I'm afraid to look at - said I'd work 40 hours/week (part time was my preference) It is actually hilarious when you think about it, all my fault, and mostly harmless, but still, thanks Claude. What this says is there's still a roll for a human-in-the-loop because AI doesn't know what matters. If you're automating with real-world consequences - job apps, for instance - maybe think about adding checkpoints. In my role as an uninfluencer, I think it's important to report both the highs and lows of my AI journey, which as been like Christmas everyday - including this one.