If you’re heavily invested in Apple apps (iMessage/Calendar/Reminders/Notes), you need a Mac to give the agent tools to interact with these apps. I think that combined with the form factor, price, and power consumption, makes it an ideal candidate.
If you’re heavily invested in Windows, then you’d probably go for a small x86 PC.
Some of those connectors are only available on the mac and some only on the iPhone. Like notes is available on the mac, but not on the phone. Vice versa for reminders.
I used Claude Code (CC) to make my own MCPs for these apps. I gave it read/write access only, no ability to delete. Of course it could probably code it's way into doing that since it can access the MCP code. I don't run it in --yolo mode though.
I interact only with CC on the machine and watch what its doing, I haven't tried OpenClaw yet.
Here's some workflows I've personally found valuable:
- I have it read the "Grocery" Reminders list and find things I commonly buy every week and pre-populate the grocery list as a starting point. It only adds items that I haven't already added via Siri as the week goes on. For example, I might notice I've run out of cheese and I'll say "Hey Siri, add cheese to grocery list". The list is shared via iCloud Reminders app between my spouse and I.
- Pre-CC, I wrote an OR-Tools python tool for "solving" the parenting time calendar. My ex and I work inconsistent schedules each month. Each month I was manually creating a calendar honoring requests, hard constraints, and attempting to balance custody 50/50. CC uses the MCPs to fetch the calendar events and review emails related to planning. It then structures everything as JSON as inputs to the optimizer. The optimizer runs with these inputs and spits out a few "solutions". I review the candidate solutions and select one. CC uses the MCP to add the solution to the calendar. This one saves me probably an hour every month.
- CC uses an email MCP to fetch emails from my child's school and suggest events its found in the emails to add to the calendar.
None of these are huge time savings on their own but the accumulation of reducing the time to complete these repetitive tasks has been awesome in my opinion. These are all things that most definitely would not have been worth automating with traditional dev work but since I can just dictate to CC for a few seconds and it has something that works a few minutes later it's become worthwhile.
I guess what’s wrong with it? Let’s say it has read only access, new messages and calendar invites need approval. I’m not sure I understand the harm? I suppose data exfiltration, but like you could start with an allowlist approach. So the first few uses and reads take a while with allowing the ai to read stuff , but it doesn’t seem that crazy given it’s what we basically do with ai coding tools?
Might be an interesting problem for understanding how various models perform recollection of prior tokens within the context window. I'm sure they could list animals until their window is full but what I'm not sure of is how much of the window they could fill without repeating.
I guess it could be generalized to filling up the context window with any token, but just making sure none of the tokens repeat.
An interesting twist could be making sure a specific token is an anagram of the token N tokens back. This could possibly measure how much a model can actually plan forwards.
Even more interesting is if a thinking LLM would come up with tricks mitigating its own known limits - like listing animals in alphabetical order, or launching a shell/interpreter with a list that contains previous answers (which it then checks each new answer against).
I really want to get into these Lora based mesh tools but the range in my experience is terrible. Maybe I'm doing something wrong, maybe it's a lack of nodes in my area.
I just tested the other day. I'm in the midwest US so it's winter, no leaves. I managed to get about a quarter mile before my two portable nodes couldn't talk to each other. T-Echo with muziworks whip antenna.
Without a bunch of solidly placed, high elevation, high gain antenna nodes, this just isn't really that usable.
Plus, all the other issues others have highlighted.
I still think you're going to be in manual threshold tuning for quite a while. The cost of feeding a continuous log to an LLM would be insane. Even if you batched until you filled a context window.
> It chews through tokens. If you're on a metered API plan I would avoid it. I've spent $300+ on this just in the last 2 days, doing what I perceived to be fairly basic tasks.
Didn’t Anthropic make it so you can’t use your Claude Code Pro/Max with other tools? Has anyone experienced a block because of that policy while using this tool?
Also really curious what kind of tasks ran up $300 in 2 days? Definitely believe it’s possible. Just curious.
Seen a couple of people on X have posted about their Claude accounts being suspended after using this. All of them seem to have used it with Claude Code so yes looks like it violates their policy (not surprising really, it breaks their TOS).
I've tried it on Codex (ChatGPT Pro) and within an hour of just getting stuff set up and tested used half my weekly limit so I can see using $300 in a couple of days being very easy.
Until thats figured out this is basically a non starter, you can't use it if its going to cost $1k+ per week to use, and I'm not sure theres any local models that'd handle it without $10k+ in hardware costs.
I’ve been working on adapting Claude Code to do some repetitive “personal assistant” type tasks so I was really excited to try this tool.
One of my tasks is a skill that fetches my calendar via MCP and slots events into a JSON to be used for an OR-Tools constraint optimizer that finds a workable schedule for something. It then uploads those events to the calendar using MCP when I choose my favorite candidate solution.
I checked token usage for this task last time I ran it. It would’ve cost $29 in API usage with Opus 4.5.
So yea, you’re absolutely right that this stuff isn’t going to go mainstream at these rates.
One thing you can try is powering Clawdbot with a local model. My company recently wrote[0] about it.
Unclear what kind of quality you'll get out of it, but since the tokens are all local, kinda doesn't matter if it burns through 10x more for the same outcome.
I offhandedly set it up to do a weather alert every 4 hours during the big winter storm. Absent a well-specified API, I can only assume it was repeatedly doing a bunch of work to access some open API it discovered.
Very much the LLM equivalent of “to bake an apple pie you must first invent the universe”.
How does it bridge iMessage? I see clawdbot is using imsg rpc on a Mac but really curious about running this stuff on an old iPhone for access to iCloud things. I have a few of them laying around so I could get started way faster.
I imagine in a situation like Iran, carrying a backpack full of WiFi gear to stay connected to the meshnet is a red flag.
Establishing a bunch of base stations is likely to raise red flags too.
It's pretty trivial for a nation-state that is jamming GPS to go around and jam WiFi or analyze WiFi spectrum for a meshnet operating in and around a protest area.
- Potential job loss, particularly in the bottom half or so of jobs.
- Further wealth inequality due to so many factors but primarily because the companies providing these tools will capture the dollars that would’ve been spent on the jobs mentioned above.
- NIMBY-ism. AI = data centers and people are overwhelmingly deciding they don’t want these near their homes. I live in the Midwest and it’s been amazing how much opposition has been showing up for these projects.
Of course all of these are based on the speculation and “promises” of the tech. Many feel the time is to act now rather than once it’s too late, on the off chance these things do happen.
>- Potential job loss, particularly in the bottom half or so of jobs.
"the bottom half" of desk jobs, maybe. But most jobs in "the bottom half" overall are not desk jobs, and therefore aren't going to be replaced with AI anytime soon. Think burger flippers, waiters, and retail clerks.
My gut says yes and no. Generally I think jobs pushing atoms are probably safer but there's been headway on this as well. Maybe not total elimination but further skill and human labor intensive reduction seems to be the trend. For example: a fast food restaurant maybe needed 5 employees before and with these advances needs 2. Things like automated order kiosks, automatic fry machines, or bots that restock pre-packed inventory. More labor intensive tasks though like plumbing, electrical, etc already pay somewhere around the 50% mark anyway because of a lack of skilled workers.
On the physical side of jobs, robotics definitely has heavy investment and I think it will continuously and slowly eat away at some of these physical tasks. Look at a modern warehouse for example.
>Things like automated order kiosks, automatic fry machines, or bots that restock pre-packed inventory.
>On the physical side of jobs, robotics definitely has heavy investment and I think it will continuously and slowly eat away at some of these physical tasks. Look at a modern warehouse for example.
None of those are "AI", though. At best it's "tech".
Don’t disagree. It’s no different than how the lay person doesn’t differentiate between LLMs (the new thing) and ML (the thing that’s been around awhile). It’s all just AI.
Blame the marketing and tech leaders who throw “AI” in literally every marketing copy produced in the last 24 months.
The US seems to m mostly look down on blue collar work, service workers and other non-desk jobs. Maybe that's part of the reason why AI as a threat to low-skilled desk work is seen as such an offense. Those affected might slide into the "lower class" of people who use their hands to earn an income, and those in the "lower class" will have a harder time climbing up into a desk job
I wonder how many of those desk jobs actually create value though.
Some paper pushing asshole working for the government demands some paper bullshit. Some other paper pushing asshole working for bigco produces said paper. Is value actually created? Perhaps there's some risk mitigation but enough to justify their respective wages? And the need to push that paper back and forth locks the little guy out of competing in that market.
Yeah, it'll suck for a lot of people in the interim. But that will also put downward price pressure on a ton of things who's cost makes other value producing things not worth doing. If legal, design, engineering, etc, etc, services are made cheap in the "boring" cases then that becomes competitive advantage for the buyers which over time trickles down to their buyers and their buyers.
>I wonder how many of those desk jobs actually create value though.
>Some paper pushing asshole working for the government demands some paper bullshit. Some other paper pushing asshole working for bigco produces said paper. Is value actually created? Perhaps there's some risk mitigation but enough to justify their respective wages? And the need to push that paper back and forth locks the little guy out of competing in that market.
Probably a minuscule (<5%) amount? Think about it. In a 100 person tech company, how many people do you think is doing legal/compliance/security? More than 5?
If you’re heavily invested in Windows, then you’d probably go for a small x86 PC.
reply