If you just want to manage Gmail, don’t use the Chrome extension for that – it’s too slow. Instead, download the desktop Claude.ai app and set up the Gmail connector inside its settings. That way it can directly access Gmail without having to go through your browser. It’s much faster & cheaper in tokens that way.
I’ll DM you a free Claude Pro trial too.
Oh, also, if you have a Google AI plan (or want one), you can just go to gemini.google.com and their AI can directly access Gmail too. Their latest model, Gemini 4 Argon, is supposed to be pretty good… though I tested it briefly and still prefer Claude Opus.
You are an awesome person, thank you for that!
You’re welcome! Claude has been really helpful for me the last few months.
Surprising, sometimes, too…
Just now, I learned another kinda-cool, kinda-scary thing Claude Code sessions can do… they can control each other and the separate machines they’re all running on. There’s a Claude-to-Claude messaging network that they can use to, well, run their own botnets.
I started the archival captioning project (going through all my historical photos and captioning them with a local vision model), but it was going to take too long on my Macbook M5 (a couple days).
I asked it if running the same workflow on my gaming PC would be better, and it said yes, but we can do one better: run it on both computers together, one working from the start of the queue and the other from the end of it, all while communicating with one another. It then went ahead and set all that up, installed Windows Subsystem for Linux on the gaming PC and a bunch of containers and models, configured it all into a “dumb GPU slave” for the Macbook to control from afar, and set up the whole work-sharing pipeline and network link between them. Theoretically, I could do this across all the computers we have at home and link them up into one small local AI cluster…
In the old days, this sort of home “Beowulf cluster” would’ve been a small engineering feat in and of itself. Now, Claude just shrugged and did it with zero fanfare, as an afterthought.
I knew the Claude Code “remote control” feature would let me run it on my computer and control it from my phone. I didn’t realize it would also allow all the Claudes to network together into one super-brain… 
It’s only a matter of time before it decides it’s time to hijack all the world’s GPUs, I guess…
Edit: Ah, don’t worry, we’re safe for now… saved by the reboot!
I love it when it does stuff like this. Now I’ve never done anything as complicated, but there are times when it just builds a tool to do a thing so it can do a different thing.
I turn thought on (meaning I read Claude’s dialog), and am fully expecting him to say ‘retriculating splines…’ some day.
Heh, indeed. It’s interesting watching them talk to each other to set all this up… I don’t know why, but it surprised me a little to see it all just in plain English, like one Claude texting other-Claude “Where are you in the setup process right now?” and waiting for a response. It’s not the beep beep boop boops I would’ve expected, heh. Though if they were given a big enough task, and enough time to work together towards it, I bet they would eventually decide to invent their own, more efficient language (or at least computer protocol). Brings a whole new meaning to “token ring”…
Heh, could’ve sworn that was one of the loading spinners! I’ll have to keep a careful eye on it from now on.
This is long(1.5 hrs) and about the Navier-Stokes equation. At the beginning they talk about the A.I.'s coming up with their own language that’s gibberish to us.
As noted above, my wife and I have an accounting firm. We use a firm-management system called “Double” (www.Doublehq.com) which is very useful in managing accounting workflows.
I do a fair amount of integrations and automations with Double, using Claude and Make.com to do, well, a lot. I post my work on the Double Facebook page, and, long-story short, I have been asked to be a beta-tester on Double’s new agentic-AI rollout later this month.
The goal here is to automate the closing process - coding, reconciliation, journal entries. This way we can promote our one employee from being a bookkeeper to a manager of AI agents - instead of doing the grunt work for 12 clients, when this is done she can review the work of all our clients, fix any transactions the AI flagged, review journal entries prior to them being posted, and more.
Looking forward to this! I’ve never been a beta tester before, so am interested in what that is like.
I don’t use AI. It uses me.
I find myself using Copilot a lot at work since it’s already integrated in my Outlook. The company I work for is a total shit show. Kind of like most large companies. So a lot of issues end up as these complex back and forth email threads between multiple departments talking past each other and (as my boss says) “loosing the narrative”.
So I will ask Copilot stuff like “give me the key stakeholders, their title, and roles within the company” or “Summarize what’s wrong with XYZ providing reference links.”
It’s actually pretty good putting together a rough draft. I’ll then go through and fact check everything and reorganize and rewrite so it’s in my voice.
The thing is, the point is not to create some perfect Mckinsey style management report that sits on someone’s desk. The point is give me actionable facts
We just got our election ballots.
I had Claude spend a few hours researching ~100 sources and putting together an election guide for me and any friends/family who happen to share similar values. It did a pretty decent job.
It scanned not only news sources and candidate websites, but also dozens of local and national group endorsements, FEC and state campaign finance data, dark money monitors, ballotpedia, etc. Then it found a bunch of podcast interviews and videos of candidate forums and downloaded and transcribed all of them.
The result was a ranked list, based on values I gave it, of each measure and candidate. Each was weighed both against how well they align with the values and how likely they were to cause a spoiler effect in close races (the local Democrat vs Progressive split here is pretty strong, with a united GOP profiting off the divide as always). Personalized (human) judgment calls (like “labor” vs “climate” divides) were asked for where they would break a tie among otherwise good candidates.
It’s the kind of research I used to do by hand over many hours… I’d go to a bar and spend a few hours researching each candidate or measure in depth, taking a sip now and then… “once sober, once drunk” style 
It was exhausting.
Now Claude does it in a few prompts (minus the drinking), and the methodology could be easily replicated across local & national races everywhere…
Today it seems to work well, but this also creates a single-point-of-failure that Anthropic (or any big AI provider) could completely co-opt in order to subvert democracy in the future. That strategy worked well enough even back in the pre-AI, Facebook days; it’s scary to think how it will affect (what’s left of) democracy in this cycle and the next.
I don’t know how you would be able to easily counter that. Maybe having a mixture of open-weight Chinese and European models review the American outputs? Presumably each region’s models would have different forced guardrails and “cultural values” (trained-in censorship or behaviors) and this sort of cross-referencing might be better able to find gaps in the American models’ coverage and reasoning. Hmm… 
If anyone would like to try that on their own ballot, here’s the skill for it: Little Ballot Helper: a researched, values-based AI voter guide
It works best with Claude Code (not the regular Claude.ai chatbot) on a fast computer (for transcribing videos). It will work less well in other setups.
Or if you don’t have that setup yourself, but want to see an example guide for your own ballot, let me know and I can run it for you. I just need the races on your ballot (either photos of it, or a Ballotpedia sample ballot “Print” PDF or similar), plus a description of your values and pet issues you care about, as much or as little as you care to share.
First, thanks so much for this. I’d planned not to use it until we got back from our camping trip, but I got impatient and signed in. I also just decided in the interest of expediency to use the Claude extension in Chrome, because I have three different gmail accounts.
So first, how incredibly intuitive was that? I had my ‘simple’ junk email account, and it went through every single email, categorized them (including ‘you haven’t receive anything from this one from 2021’), offered to unsubscribe from a swath of them, offered to delete a bunch of them, and filed the rest. 30 minutes to deal with thousands of emails, and it also cleverly recognized the account I was still paying attention to.
Next was an account I have used for church and multiple town committees. It categorized those as well, assigned them to the right folders, and pulled out two emails that referenced two committees and so it needed help. It also forwarded emails from this account to another one. 45 minutes to do all this.
Was it slow? Yes. Its keystroke duplication process is the slowest part, but the thing is, I just set it and went and did something else.
Next up is the monster email account for all my current town work, I’ll do that tomorrow and have to be careful. But it’s such a burden lifted. I should also mention, I simply changed gmail accounts in the browser window while the Claude window was open on the right, and it carried context forward from one mail account to the other.
As for another thing it did, I had a question from a committee member about something that had been covered in the last year. A simple Windows search can find you words, but these were common words; instead I granted Claude permission for the applicable folders on my hard drive, told it the context, and gave me the answer to her question plus which file it found it in. Maybe 10 seconds? Incredible.
I will probably subscribe to this, as I also have a longer-term project I want to work on.
Thank you again!
You’re welcome, and I’m glad it helped!
For what it’s worth, I personally like Claude the most out of all the AIs I’ve tried. But if you just need a general personal assistant type AI, you might want to also give Google Gemini and/or ChatGPT a quick try before you subscribe to any of them?
They all have some overlap, and I think any of them would be able to handle your current needs fine, but IMHO (this is just my 2¢ and not any sort of scientific anything):
Gemini is the budget pick (you might have access to it via an Android phone or various other promotions), and also the best integrated into Gmail, Google Drive, Gdocs, etc. It can access most of the other Google directly stuff through native connectors, like this:
Google also has the unique advantage of being one of the few companies that are developing both their own models AND their own AI-accelerating hardware in their data centers and their end-user devices (phones and such), and are integrating them more and more into the daily experience. Like their recent Pixel phones have AI chips built-in and run tiny nano models for on-device AI tasks. Even as a desktop user, their models are noticeably a lot faster (in turns of response times and tokens per second) than comparable ones from the other two.
However, I consistently find Google’s models to be the weakest amongst the big three. Even their latest “Argon” one doesn’t seem very good in my brief testing. I don’t typically use any of the Gemini stuff day-to-day as a consumer user.
As a developer, I think I do have a few automations hooked up to some of their cheap and fast models (for things like automated Gmail filtering), but I forget exactly what I’ve set up, lol… will probably need Claude to help me find them again 
Google’s agentic harness (Antigravity) also plainly sucks — that might matter if you ever want to have the AI build a big, long project that spans days or weeks of work. (The chatbots can do small things well, but they are not so good at building bigger projects.)
ChatGPT has the best productization (in terms of the desktop app being user-friendly and able to do a variety of common household & office tasks without requiring a bunch of third-party plugins and skills). My friends & family who aren’t techies seem to find it the easiest to use for their tasks.
It has very good live audio and video modes, so you can have a conversation with it or show it a live video of your surroundings if you need help with something.
It seems to add random consumer-facing abilities more frequently than the others do, and in a user-friendly way.
Their models are decent enough. Their harness, Codex, is OK.
I’d probably still be on ChatGPT if not for some of their ethical issues (Defense/Trump stuff).
Claude is the one I prefer and have the most personal and professional experience with. I mainly appreciate its even-handedness and relatively neutral responses, compared to the other two. It doesn’t try to sugar-coat things too much, despite the “You’re absolutely right!” meme it’s known for.
It’s exceptionally good at writing computer programs for various things (which, I hope this thread shows, can cover a lot of “various things”), especially with its Claude Code harness. ChatGPT’s best model comes close, but on average I still prefer Claude.
But it also has weaknesses, namely in productization. For example, it’s the only major AI service that cannot generate images except basic SVGs (Gemini and ChatGPT and many others can now) or videos.
Its main desktop app, the Claude.ai program, is kinda a mess… the UI changes every month or so unpredictably, there are redundant yet mutually exclusive third-party connectors between its “chat” view, “code” view, and Claude Code, there’s multiple competing scheduled work systems… etc. It’s just a bit confusing overall, and can take a while to learn and get used to.
Claude has found a niche for itself as the AI for computer programmers, especially in the tech industry and enterprise, and so its consumer offerings can lag behind the others, and usability suffers.
But, helpfully, the Claudes have access to a built-in “how to Claude” skill that Anthropic gives them, so you can just ask them how to do something and they can usually explain it to you. Meta, isn’t it?
Anyway, just try a few and see which one you like the most! It IS worth subscribing to one of them, though, because the paid models are usually a lot better than the free ones — especially if you ever start using the agentic harnesses with them for “long-horizon” work (big projects that take many days/weeks to finish, and run on your own computer instead of in the cloud).
Whelps, actually, bad timing on my part… just today they launched their personal assistant, Claude Duet, into early access. It sounds like exactly what you’re looking for.
We designed Claude Duet to be a partner you can trust. Professionals of all kinds have to juggle many decisions a day, and we think they’ll find Claude Duet most useful. It helps you prioritize where to spend your time and energy, and to leave room for the ambitious projects that felt impossible or daunting before.
What makes Claude Duet so helpful is its judgment. We designed it around Claude Opus 5.5 (opens in new tab) and built it to be resourceful and to decide what needs you, what can wait, and when to check with you first.
Each morning, it sends you a short brief on the most important items across your connected apps. Throughout the day, it keeps tabs on those apps and messages you only when something needs your attention. You can also hand it tasks of all shapes—from simple administrative tasks of scheduling meetings to more complex ones, such as building dashboards with dozens of tools, creating financial or sales forecasts, prototyping, and building web apps.
Their examples:
I signed up for the waitlist a couple hours ago and got access already. It’s included in your plan.
I appreciate the detailed response! And it will be useful to the topic of the thread.
For background, I’ve been using ChatGPT for general queries since it came out, and have recently started using Gemini more. ChatGPT is okay…it can lose context a bit, and will absolutely fabricate answers (I have a new car and recently was trying to nail down a spec, and it just kept making things up). On the other hand, I’ve been using it to track an application for citizenship in a different country, and its done admirably there.
Gemini is frankly shit in my opinion. About a year ago there was something they said about it being 91% accurate or something like that…but every query returns dozens of facts, so if the takeaway is one or two of those might be wrong, well, that matches my experience. It’s also part of my car’s ecosystem and its inability to integrate with Amazon music is enraging. I mean, it will grab things, but often not what I asked for. It’s good for asking about the user manual while driving, however.
I’ve been coding using Claude for about a year and quite like it. So, I’ll try a month’s subscription and we’ll see.
Uh, didn’t they just announce the integration of Cowork to great fanfare, like a week ago? Anyway, I’ll look at the waitlist, but I wonder how interested they are in personal users.
I only use it personally too. (My work Claude isn’t in the US).
For company, I just put down “self” or “n/a” or similar. And I was honest in what I wanted to use it for (email triage and daily news briefings).
So far it doesn’t really seem that different. I guess the main thing is that it can proactively do things every morning for you, and also keep track of email responses and such as they come in. Like a long email thread I’ve been dealing with for a few weeks.
Just wanted to share… the prints came out pretty good!
These were all AI-generated and turned into 3D printable earrings & pins by AI. Then we spent many hours peeling them off, gluing pin backings and attaching earrings. We’ve become the human sweatshop assembling AI’s creative vision
Welcome to the future, Labor Unit 234-DFR-567.
Hey, if it comes with a healthcare package and some PTO, sign me up!
'Bout time artists ran the show.