How are people setting up virtual assistants?
From the mailbag
Today is our last mailbag reply for this round (our first anniversary edition). We’ll open it up again in June for another chance to submit your AI questions, confessions, and use cases.
Submission: Everyone seems to be setting up personal assistants that pay parking tickets and give them feedback on how they're performing against goals. How does one do this?
Let's break this one up into parts.
How are people getting AI to pay their parking tickets?
You have to move beyond the chat harness. A harness is a container for the AI. Chat is one, but Claude in Excel, Claude in Chrome, and Claude Code are also examples of harnesses, just to name some of Claude’s. To pay a parking ticket, you’d need AI to be able to navigate a graphical user interface, aka surf the web, so there’s really two options here.
One is to use a tool like Claude in Chrome. You’ll still have to enter your own payment details, but it can help copilot tasks you’d otherwise avoid or feel frustrated by. (Speaking from some experience here.) For these sorts of tasks, AI still often takes the same amount of time or is slower than doing it yourself, the baseline benefit is that you don’t have to do it yourself. Where it adds value is when you run into complexities or ambiguities. For example, if you’d like to argue that parking ticket, AI could help you write an effective complaint and figure out where & how to submit it.
If AI was fully paying the ticket, the person was probably using OpenClaw. OpenClaw has big enough security holes that even I, who have made much of my life an AI experiment at this point, am fine waiting for these features to make their way to safer environments. OpenClaw has now been purchased by OpenAI (ChatGPT), and Anthropic (Claude) is swiftly implementing many of the most valuable features from OpenClaw into Cowork.
How are people getting AI to review their progress against goals?
I’m going to go ahead and assume work goals here for simplicity, but the same general flow would apply if you were talking about personal goals.
If you save your goals to memory, connect the primary places you work to your regular chat-based AI–your drive, calendar, email–and attach the review output format you are going for, AI can draft a progress report for you*. This would be the “Let AI do it for you approach,” which may or may not make sense for you & your context. There are other constellations that would work here too.
I generally like to organize my own mind first, so I’m pretty unlikely to take the above approach. I’m much more likely to collect my own thoughts, probably on paper or maybe in an Apple Note or Google Doc, then bring in AI to source specific evidence for my views and to suggest ideas I’m missing or blindspots I might have.
I’m also very likely, once I’ve worked out my preferred process, to build a skill, so I can efficiently run the process when the time comes around again. Skills encode the steps and preferences for tasks so that you don’t have to start cold the next time you come back to the task.** Claude debuted them in October and has the most advanced version of them, and ChatGPT added them in March.
So, how are people setting up virtual assistants?
For most non-technical people, the best place to do this right now is in Claude’s Cowork, which is a safer, easier-to-use version of Claude Code (albeit, with somewhat fewer features). Explaining how to use Cowork for this purpose is wayyyyy more than I can cover here, but is something we build the skills for in our Claude for Work class. I’m not holding out on you--if I could give you a more satisfying answer here, I would (and then I wouldn’t teach a class on it).
