I think it’s a great guide, and I was also on the fence. I don’t ‘trust’ AI, but David’s approach isn’t to trust it either. I think (as ever) there’s a really good course charted here that is very ‘real world’ and for the first time with my own use of AI I felt as though I was really getting some ‘help’ from my AI client.
This has all come to a shuddering halt, as Claude/Anthropic are having some pretty serious issues around limits at the moment. I’ve suggested on another thread that one of the workshops might cover how to maximise token efficiency with the robot assistant, as tokens are becoming as expensive as petroleum…
Okay reporting out here a fascinating new direction I’m taking RAFG.
I’m only half way through the FG, but far enough along that I understand skills a lot better and am using the Obsidian environment, which I’ve long been a naysayer but honestly it works great. I’m also using the Omnifocus MCP. So, here’s the use case that feels promising for my workflow:
This weekend, I spent a good bit of time doing a big GTD-style sweep and re-orientation in OF (as you do) and there are all these nicely defined tasks and projects per Claude’s organizational skills.* This morning I prompted Claude:
I'd like you to go through Omnifocus and flag any items that you think I might be able to meaningfully delegate to you. I'll review them, unflag ones I don't want help on, and then we can pick one and see if it works out.
It flagged a dozen or so tasks, I unflagged half, and then I told it I was going to go away for a while and be unavailable (I took a shower), and it went through and created six separate documents in Obsidian that scaffold my next steps. A couple of examples:
I’m developing a protocol for how we credit contributors in participatory design work. Claude researched best practices across disciplines and produced a two-level protocol design: one for all participants in our work, one for granular team self-assessment.
This was just a proof-of-concept, but I’m genuinely excited. David mentions how he wants a daily briefing like CEOs or presidents, my version of this is I’ve always wanted to be able to delegate tasks but I’ve mostly operated as an IC and this wasn’t possible.
–
* Another win here is having Claude frame tasks in actionable ways, something I’ve never been particularly good at.
Ooh, I like this! There’s a Things MCP server I’m willing to take a risk on, so I’ll have to try letting Claude look through my task list for items it might help me with.
I’ve already managed to create a skill (more accurately, have Claude create it) for adding a lesson-planning database to a Notion page of my choosing. Handy for course development and planning.
I don’t want to create any misunderstanding — I find the whole development around AI-powered automation really exciting! From my perspective, OmniFocus is a good example that puzzles me: isn’t the entire end-to-end encryption, which is a key quality feature of OmniFocus, rendered moot if I send task details to an external AI server? At that point, I’d surely need to at least pseudonymize my sensitive, privacy-oriented items that I previously wrote in plain text. That would leave me with a choice: either preserve an important feature of OmniFocus and leave the AI-advantages alone, or have my OmniFocus-data intelligently and automatically analyzed and processed but sacrifice privacy. Or am I missing something here?
I appreciate you bringing this up bc it’s a decision we all must make based on our own use and values. And our use and values interact with an enshittified ecosystem. These tools, even beloved ones, compete in an economy that gravitates toward locking us in. The more useful any of them are, the more likely we’ll be locked in at which point they can become less useful and retain us bc the switching costs are so high. I’ve recently canceled my OpenAI account and exported my history but there’s nothing I can really do with the data, or the value it had (though the same is not true for OpenAI).
As for E2E encryption, this feature of OF wasn’t a selling point for me. I don’t use it in a way that I would worry too much that OG could see it. That said, it’s an important thing to point out especially for those who do.
The truth of the matter is that the nature of my interactions with Claude are far more sensitive than any task I put in OF. At my research group, we call this the “data sunken costs” (privacy already feels so forsaken that you share more with AI). I’m ambivalent about it… on the one hand concerned with the privacy I’m giving up to find use in these tools, and on the other hand needing to learn how to use them well to keep up - another term we use “expectation explosion” (when AI-driven efficiency leads to the expectation to produce more, faster, better, raising both what is possible and what is expected).
You can use that MCP with a local model, too, to keep it all on device. The extended writing/research quality won’t be quite as good, but Qwen (for example) would parse the tasks just fine.
This is exactly my dilemma. As I work through the Guide, thinking about how I might use the techniques, I am also working through how I might protect my data and privacy.
Some things Claude et al can do, I’d rather just Perl for, honestly.
I’ve had a rocky two weeks with the daily brief losing track of things and interacting with BusyCal via MCP. I THINK (he said hopefully) that we are now past that. The biggest issue I have is getting it to work reliably witgh Chrome. Still working on that.
OTOH, Claude did build a photo clubstatic site, and populate it with photo ;locations in OR and NM
And the meeting summaries with key poiunts and actions - which are then uploaded to my perrsonal calenar or company Notion seem to be working well.
I bought the robot guide, even though I didn’t have any specific uses for it. So far it’s been a lot of interesting information and I really enjoyed going through the parts that I’ve gotten through so far, but the honest truth is I haven’t had this much fun playing with my computer in well, I can’t remember how long. My problem is this is a rabbit hole that I could make super super deep, I’m having a blast.
I agree this is a book that’s entertaining. Daniel Suarez has several novels where technology provides the background for the adventure. Daemon is a good place to start becoming familiar with his work.
