Hi James, thank you for the detailed analysis. I am trying to implement the same thing myself. I will definitely take inspiration from all your features.
I saw something that interested me more than anything else: Omnifocus's MCP.
I switched to Todoist for the claude connection. With Omnifocus's MCP, however, I could immediately go back to Omnifocus (an app I love for tasks).
If you’re ok with using the desktop, you get it out of the box.
I further wrapped it in a server that made it available to Claude on the web (so mobile app or Claude.ai website) and used a Google OAuth flow. Note: this IS the local Omnifocus MCP, I just have it permanently running/open on my Mac Studio.
It’s not particularly complicated, though it may be better to actually use Claude Code locally, use a remote session (or tmux), to use OmniFocus on your running computer if you aren’t comfortable wrapping it/setting up OAuth. You need your computer running with OmniFocus running either way.
Incredible setup! And thanks for the shout-out. I'? definitely a lot of inspiration from this article to improve my own workflows. The cron integration is something I've been thinking about a lot lately and now you gave me the final push I needed to try it out. Good job, man!
Thanks! Tactically, if you’re on MacOS, you’ll actually want to put it on launchd if you’re using Claude’s subscription (based on how Anthropic stores their auth token). Just as an FYI, which took me a good 30 minutes to figure out at one point.
Separately, we need to do a refresh on that interview both to actually have a post-able one + to update on everything that’s happened!
You definitely can depending on your stack! I’m a bit of a computer hoarder, so have a bunch of Mac Minis and Mac Studios. I really need to pair down, but in the meantime I try to utilize them as much as possible (not that this strains it much). I also pay for 10Gb fiber so… Separately, DevonThink and OmniFocus are Mac only, so that’s another constraint. Mac VPS is ridiculously expensive.
I’ve been experimenting with running very small local models on some of them, but it just isn’t really worth it based on pricing. I only really use local models like Whisper for stuff that’s 1) very very sensitive, 2) has relatively clear audio quality.
I also use Eero so have their Dynamic DNS stuff to point to my home gateway. But I used to use Cloudflare’s API and before that DuckDNS.
Great piece, really appreciate your calling out the rewards as well as the risks. As you note YOLO is where the power is at, and obviously creates risks like lethal trifecta.
What do you think in the short term “disruption” might look like from someone who sounds like closer to optimistic “Fully Automated Luxury Gay Space Communism” view of AI.
Let’s say most tech companies implement a Dorsey-like downsizing of 50% of white collar workers and there’s a recession fueled in part by this “primary source” of unemployment followed by “secondary unemployment” because of the reduced consumer demand of restaurants, travel/entertainment, personal services, etc.
Do you expect the political system to respond or to worsen the crisis by lack of effective response (like the first year of COVID)?
I take it the FALGSC folks think things will just adjust through the magic of capitalism that despite the temporary pain of transition, things will bounce back bigger than ever with new industries, occupations, opportunities etc.
Do you? What would that transition look like, especially for those unemployed by the changes.
And lastly, many of the believers in automation hold out the utopian ideas that the huge increases in productivity will “trickle down” to non tech, non STEM “agentic” workers in the form of a shorter workweek, more leisure, universal basic income, etc.
Yet the development of capitalism including tech in the U.S. is to capture the almost all productivity gains for the 0.01% creating billionaires and are not distributed workers wages or benefits, except for a few lucky superstars like perhaps OP. Most businesses seem to want to ruthlessly reduce labor costs and eliminate head count.
Given current trends how can you remain sanguine that the “painful disruption” will be short and have a relatively happy ending? Is there anything which could be done to reduce the pain, and what’s the likelihood that would happen?
I’m pessimistic. If unemployment shoots up to double digits and the answer is “patience”, what does that look like in five or ten years?
Yeah, I’m not in the FALGSC camp. I don't think I'm an optimistic enough person generally, lol. Anyway, I don't think things bounce back through the magic of capitalism. My argument is that the mechanism of disruption follows historical patterns (mechanical skills devalued, judgment skills elevated, fields grow even as specific roles shrink). That’s different from saying the transition is painless or that benefits distribute fairly.
On inequality—I’d push back a bit on the framing that productivity gains go almost entirely to the 0.01%. The wealth concentration story at the very top is real, but that’s largely an asset appreciation story (stock prices, platform economics). The wage story is more nuanced. Real wages at the bottom have actually grown meaningfully in recent years, especially post-COVID—the Fed wage tracker has consistently shown stronger growth for lower-income workers. So the true version of the concern isn’t “all gains go to billionaires.” It’s “will labor markets stay tight enough for workers to capture some of the surplus?” And that depends on whether AI causes mass displacement or gradual role evolution. I've talked about before in other pieces that AI works far better in complement with humans.
On the Dorsey scenario—I actually think the saving grace is adoption is way slower than people fear. That NBER study of 6,000 CEOs finding 90% reporting zero AI impact on employment isn’t because AI doesn’t work. Reality is, organizations are slow. The more likely path isn’t a coordinated 50% layoff wave but slower trickle: slower hiring, attrition not backfilled, role consolidation. Still painful for those affected, but not the sharp demand shock you’re describing (or Citrini's thought experiment).
Besides that... if I'm being totally honest, most big tech companies ARE bloated. My personal opinion is a lot of these flashy layoffs are using AI as an excuse.
That being said—and this is the part where I’m pessimistic—the thing that actually worries me isn’t the big headline layoffs. It’s what I’m seeing anecdotally across small startups, biotech companies, hedge funds, and similar. Small businesses just… not hiring young ppl. A senior person with AI tools decides they don’t need that junior hire. That’s roughly half the private workforce in the US. that also doesn't show up as a big, dramatic splash. It shows up quietly as a generation that never gets on the ladder.
And look, my own argument that AI’s biggest benefits accrue to experienced experts who can act as “oracles” depends on people actually getting that experience somewhere. If nobody’s investing in juniors, the very thing that makes AI productive—human expert oversight—gets undermined over time.
So no, my answer isn’t “patience.” I share some of your pessimism about the political system responding well. Our politicians really haven't covered themselves in glory. Another piece I'm noodling on is I think regulation is important... yet I think it's better not to have any... because I think our politicians suck so much they'll screw it up. That's a root cause of a broken political system though, which AI can't solve (and doesn't cause).
Anyway, the macro displacement story probably follows historical patterns. The generational pipeline story is the one that worries, and I don’t think we’re talking about it enough. I do think it WILL resolve (architects figured out how to employ junior architects who were no longer hired on the ability to draw straight pencil lines after CAD came along... eventually), but that's where more of the jarring stuff could be.
So, taking architecture as an example, was there an effect on the pipeline for many years where higher education was looked at as a mugs game and people avoided entering the field due to the employment prospect collapse? Did that problem just sort of quietly resolve itself and come back into equilibrium after a generation or two? Or did it become an economically unrewarding “passion occupation” similar to “helping professions” or non-profits?
Complicated. To some degree, architecture was always a bit of a rich kid’s profession (like acting and journalism). My point is more it wasn’t “there were never junior architects again, the end.” It’s a neater example than most “professions” which in reality for most regular people is way more fluid and less defined. For some of this macro + politics stuff, you should definitely also follow Paul Podolsky.
Thank you for sharing your setup, very thought-provoking! Good luck with your book launch
Cool stuff. You're a brave man. Thanks for sharing the details.
Hi James, thank you for the detailed analysis. I am trying to implement the same thing myself. I will definitely take inspiration from all your features.
I saw something that interested me more than anything else: Omnifocus's MCP.
I switched to Todoist for the claude connection. With Omnifocus's MCP, however, I could immediately go back to Omnifocus (an app I love for tasks).
What did you use for Omnifocus?
Thanks
This one (it’s the one that comes up if you search for Omnifocus in Claude’s connectors—no need to go to GitHub). https://github.com/themotionmachine/OmniFocus-MCP
If you’re ok with using the desktop, you get it out of the box.
I further wrapped it in a server that made it available to Claude on the web (so mobile app or Claude.ai website) and used a Google OAuth flow. Note: this IS the local Omnifocus MCP, I just have it permanently running/open on my Mac Studio.
It’s not particularly complicated, though it may be better to actually use Claude Code locally, use a remote session (or tmux), to use OmniFocus on your running computer if you aren’t comfortable wrapping it/setting up OAuth. You need your computer running with OmniFocus running either way.
Incredible setup! And thanks for the shout-out. I'? definitely a lot of inspiration from this article to improve my own workflows. The cron integration is something I've been thinking about a lot lately and now you gave me the final push I needed to try it out. Good job, man!
Thanks! Tactically, if you’re on MacOS, you’ll actually want to put it on launchd if you’re using Claude’s subscription (based on how Anthropic stores their auth token). Just as an FYI, which took me a good 30 minutes to figure out at one point.
Separately, we need to do a refresh on that interview both to actually have a post-able one + to update on everything that’s happened!
Do you want to guest post this on AI made simple. I get a lot of requests for posts like this and this is probably as good as something I could write
Yeah, happy to do so! Let me know how you want to do it
Thank you for the rundown James! Are you running all of this from a VPS provider by any chance? Cheers.
You definitely can depending on your stack! I’m a bit of a computer hoarder, so have a bunch of Mac Minis and Mac Studios. I really need to pair down, but in the meantime I try to utilize them as much as possible (not that this strains it much). I also pay for 10Gb fiber so… Separately, DevonThink and OmniFocus are Mac only, so that’s another constraint. Mac VPS is ridiculously expensive.
I’ve been experimenting with running very small local models on some of them, but it just isn’t really worth it based on pricing. I only really use local models like Whisper for stuff that’s 1) very very sensitive, 2) has relatively clear audio quality.
I also use Eero so have their Dynamic DNS stuff to point to my home gateway. But I used to use Cloudflare’s API and before that DuckDNS.
Great piece, really appreciate your calling out the rewards as well as the risks. As you note YOLO is where the power is at, and obviously creates risks like lethal trifecta.
To mitigate, we built an agent harness with a policy engine that uses policy-as-code to give deterministic rules to agents. We have our SDKs up here: https://github.com/sondera-ai/sondera-harness-python
And for coding agents as well: https://github.com/sondera-ai/sondera-coding-agent-hooks
Can read about how we applied it to OpenClaw as a research extension here: https://securetrajectories.substack.com/p/openclaw-rm-rf-policy-as-code
What do you think in the short term “disruption” might look like from someone who sounds like closer to optimistic “Fully Automated Luxury Gay Space Communism” view of AI.
Let’s say most tech companies implement a Dorsey-like downsizing of 50% of white collar workers and there’s a recession fueled in part by this “primary source” of unemployment followed by “secondary unemployment” because of the reduced consumer demand of restaurants, travel/entertainment, personal services, etc.
Do you expect the political system to respond or to worsen the crisis by lack of effective response (like the first year of COVID)?
I take it the FALGSC folks think things will just adjust through the magic of capitalism that despite the temporary pain of transition, things will bounce back bigger than ever with new industries, occupations, opportunities etc.
Do you? What would that transition look like, especially for those unemployed by the changes.
And lastly, many of the believers in automation hold out the utopian ideas that the huge increases in productivity will “trickle down” to non tech, non STEM “agentic” workers in the form of a shorter workweek, more leisure, universal basic income, etc.
Yet the development of capitalism including tech in the U.S. is to capture the almost all productivity gains for the 0.01% creating billionaires and are not distributed workers wages or benefits, except for a few lucky superstars like perhaps OP. Most businesses seem to want to ruthlessly reduce labor costs and eliminate head count.
Given current trends how can you remain sanguine that the “painful disruption” will be short and have a relatively happy ending? Is there anything which could be done to reduce the pain, and what’s the likelihood that would happen?
I’m pessimistic. If unemployment shoots up to double digits and the answer is “patience”, what does that look like in five or ten years?
Thanks for the comment.
Yeah, I’m not in the FALGSC camp. I don't think I'm an optimistic enough person generally, lol. Anyway, I don't think things bounce back through the magic of capitalism. My argument is that the mechanism of disruption follows historical patterns (mechanical skills devalued, judgment skills elevated, fields grow even as specific roles shrink). That’s different from saying the transition is painless or that benefits distribute fairly.
On inequality—I’d push back a bit on the framing that productivity gains go almost entirely to the 0.01%. The wealth concentration story at the very top is real, but that’s largely an asset appreciation story (stock prices, platform economics). The wage story is more nuanced. Real wages at the bottom have actually grown meaningfully in recent years, especially post-COVID—the Fed wage tracker has consistently shown stronger growth for lower-income workers. So the true version of the concern isn’t “all gains go to billionaires.” It’s “will labor markets stay tight enough for workers to capture some of the surplus?” And that depends on whether AI causes mass displacement or gradual role evolution. I've talked about before in other pieces that AI works far better in complement with humans.
On the Dorsey scenario—I actually think the saving grace is adoption is way slower than people fear. That NBER study of 6,000 CEOs finding 90% reporting zero AI impact on employment isn’t because AI doesn’t work. Reality is, organizations are slow. The more likely path isn’t a coordinated 50% layoff wave but slower trickle: slower hiring, attrition not backfilled, role consolidation. Still painful for those affected, but not the sharp demand shock you’re describing (or Citrini's thought experiment).
Besides that... if I'm being totally honest, most big tech companies ARE bloated. My personal opinion is a lot of these flashy layoffs are using AI as an excuse.
That being said—and this is the part where I’m pessimistic—the thing that actually worries me isn’t the big headline layoffs. It’s what I’m seeing anecdotally across small startups, biotech companies, hedge funds, and similar. Small businesses just… not hiring young ppl. A senior person with AI tools decides they don’t need that junior hire. That’s roughly half the private workforce in the US. that also doesn't show up as a big, dramatic splash. It shows up quietly as a generation that never gets on the ladder.
And look, my own argument that AI’s biggest benefits accrue to experienced experts who can act as “oracles” depends on people actually getting that experience somewhere. If nobody’s investing in juniors, the very thing that makes AI productive—human expert oversight—gets undermined over time.
So no, my answer isn’t “patience.” I share some of your pessimism about the political system responding well. Our politicians really haven't covered themselves in glory. Another piece I'm noodling on is I think regulation is important... yet I think it's better not to have any... because I think our politicians suck so much they'll screw it up. That's a root cause of a broken political system though, which AI can't solve (and doesn't cause).
Anyway, the macro displacement story probably follows historical patterns. The generational pipeline story is the one that worries, and I don’t think we’re talking about it enough. I do think it WILL resolve (architects figured out how to employ junior architects who were no longer hired on the ability to draw straight pencil lines after CAD came along... eventually), but that's where more of the jarring stuff could be.
So, taking architecture as an example, was there an effect on the pipeline for many years where higher education was looked at as a mugs game and people avoided entering the field due to the employment prospect collapse? Did that problem just sort of quietly resolve itself and come back into equilibrium after a generation or two? Or did it become an economically unrewarding “passion occupation” similar to “helping professions” or non-profits?
Complicated. To some degree, architecture was always a bit of a rich kid’s profession (like acting and journalism). My point is more it wasn’t “there were never junior architects again, the end.” It’s a neater example than most “professions” which in reality for most regular people is way more fluid and less defined. For some of this macro + politics stuff, you should definitely also follow Paul Podolsky.
https://paulpodolsky.substack.com/p/ai-is-todays-tractor
Thanks for the reply; subscribed to PP, thanks. (my demo, fyi, 76m,
retired environmental attorney still active in some salient Reddit fora).