Lenny, this was an incredible deep dive—truly one of your most comprehensive and energizing pieces yet. I love how you demystified Claude Code for non-technical readers and reframed it as an “on-device assistant” rather than just a coding tool. The sheer variety of real-world examples you curated is gold for anyone exploring AI-driven workflows.
That said, I’ve noticed that in practice, Claude can be considerably more expensive than other options like ChatGPT, Copilot, or Gemini—especially for those who lean into it heavily. Many argue it’s worth the premium (and I see why), but for users on a tighter budget, it might not always deliver the best “bang for the buck.” Still, your overall argument—that Claude Code unlocks entirely new creative and operational possibilities—is compelling and inspiring. Fantastic work, Lenny. 👏
Appreciate that, Lenny, & I totally agree. The accessibility and speed are improving fast, & I’m curious to see where the price curve lands once local & agentic use cases really scale. Thanks again for your thoughtful response.
Thanks so much Lenny for this. As a non-techie, the terminal itself is intimidating but i'm hoping this will help me get much more comfortable with this.
I have another question though, is there a similar product/feature/LLM that can sit on my android and help me manage my phone experience better? e.g Aggregating stuff from multiple apps, managing notifs etc?
I use Claude Code as my first pass architect. I have our code in GitHub (read only, they won't let me write) on my desktop and I ask Claude about features I'm thinking about and to help me draft a feature plan document. Having it look at our code base gives me better results than just using ChatGPT and this is a way of brainstorming with technical input that I'd otherwise be lacking.
I use Claude Code to edit videos by using FFmpeg. I have actually used this in conjunction with Apify MCP to provide context on my LinkedIn profile/audience & edit the videos to make them resonate - Happy to provide a video
@LennyRachitsky: I came here looking for the same solution. Both 42 and 44 are incorrect links. I'm specifically interested in "42. Running a branding studio". However it shows an incorrect URL (http://run%20a%20branding%20studio%20for%20s/)
The same goes for #44, which @MattLong mentioned here. It also has an incorrect URL
Curious about this. Is it that Claude Code has a unique ability to do the above, or is it that it the same can be achieved just as well with something like OpenAI’s models, especially as they’ve recently adopted the Agent Skills Convention?
while I'm a heavy user of claude code and love it, I'm quite surprised to see it recommended for all these non coding tasks. I'm not questioning that it works, it clearly does, but it seems very inefficient and potentially counterproductive.
As we know, a large part of what makes claude code, or any other AI tool, effective is its prompt. Claude code is optimized for coding and carries a lot of tokens and instructions in its system prompt that are hardly relevant to blog writing or image optimization.
It seems to me that for most of the use cases listed here, using claude for desktop would be far more efficient and effective (lower token counts, less context pollution, more usage/hitting window limits, better output). It has all the same benefits - it can run locally, deal with large files, has MCP and connections to tools like linear or github (and I have done most of the things in the uses cases myself with it) so no downsides other than not running in a terminal (which is a big plus for some).
Is there something I misunderstood about the recommendation? has someone concluded that claude code indeed performs better for these use cases than Claude desktop?
Good feedback, Spike! I think much of this is technically true (you can technically achieve many of the same things in Claude), but there's something about the experience of using it as a terminal-based chatbot that feels very different, and makes a lot of things described here easier. I've never imagined doing most of these things in Claude itself, and having it running like this opened up my mind to what Claude can do.
Thanks, Lenny, really enjoyed this one. Love the practical step-by-step guide to open as well.
One small clarification though, since this is aimed at non-technical folks. Where the article states Claude Code “runs locally,” the Claude model itself isn’t actually running on your computer. The model still lives in Anthropic’s cloud. What’s local is the desktop app that can read your files, manage context, and handle all that clever retrieval work.
It’s an important distinction, since otherwise people might think it’s fully offline or entirely private. In reality, only the relevant snippets are sent to the servers while the rest stays on your machine.
Still, totally agree with your main point the local integration layer is what makes it feel so fast and capable.
Really good point, Kanishka! I should have been clearer about that nuance. That that model itself isn't running locally. I'll update the post to reflect this.
This was very inspiring. I have a question about ecosystems, especially since I’m used to OpenAI’s models and API.
Is Claude Code uniquely capable of the above, or can OpenAI’s models, especially with their recent adoption of the Agent Skills Convention, achieve the same?
I want to avoid switching ecosystems unless there’s a genuine advantage to using Claude Code, in which case I’d be happy to.
Thank you so much, Lenny! This piece really shows the breadth of ways people are weaving Claude Code into their workflows from teachers drafting lesson plans to designers whipping up prototypes. I especially loved the example of marketing teams using it to generate campaigns on the fly. As a non coding PM I have built my own command center around Claude Code + Wispr Flow. Markdown is my single source of truth for everything from PRDs to roadmap decisions and I spin up purpose built agents (what I call Claude Go) to handle specific tasks like requirements checks, surfacing customer insights, suggesting test cases and aligning the roadmap. The speed and depth of insight these agents provide is incredible. I am convinced that tools like Claude Skills and the new ecosystem around it will transform how we work if we embrace them with an open mind. At Why Not Us Labs we are exploring ways to make these powerful tools accessible to everyone and I am always happy to swap notes with anyone curious about how to harness this tech 🚀🕺🚀.
Lenny, this was an incredible deep dive—truly one of your most comprehensive and energizing pieces yet. I love how you demystified Claude Code for non-technical readers and reframed it as an “on-device assistant” rather than just a coding tool. The sheer variety of real-world examples you curated is gold for anyone exploring AI-driven workflows.
That said, I’ve noticed that in practice, Claude can be considerably more expensive than other options like ChatGPT, Copilot, or Gemini—especially for those who lean into it heavily. Many argue it’s worth the premium (and I see why), but for users on a tighter budget, it might not always deliver the best “bang for the buck.” Still, your overall argument—that Claude Code unlocks entirely new creative and operational possibilities—is compelling and inspiring. Fantastic work, Lenny. 👏
Really valuable element to add. Thanks Mark.
Based on the trends, I suspect this stuff will get cheaper and more powerful, but the spend will still be meaningful.
Appreciate that, Lenny, & I totally agree. The accessibility and speed are improving fast, & I’m curious to see where the price curve lands once local & agentic use cases really scale. Thanks again for your thoughtful response.
Thanks so much Lenny for this. As a non-techie, the terminal itself is intimidating but i'm hoping this will help me get much more comfortable with this.
I have another question though, is there a similar product/feature/LLM that can sit on my android and help me manage my phone experience better? e.g Aggregating stuff from multiple apps, managing notifs etc?
Good question, not aware of anything like that.
I use Claude Code as my first pass architect. I have our code in GitHub (read only, they won't let me write) on my desktop and I ask Claude about features I'm thinking about and to help me draft a feature plan document. Having it look at our code base gives me better results than just using ChatGPT and this is a way of brainstorming with technical input that I'd otherwise be lacking.
Suuuuuuch a cool use case! Thank you for sharing!
I use Claude Code to edit videos by using FFmpeg. I have actually used this in conjunction with Apify MCP to provide context on my LinkedIn profile/audience & edit the videos to make them resonate - Happy to provide a video
Whoa! Yes, please share an example, to see what kind of videos this workflow can create.
Lenny, do you have another link for 44? The URL is broken in the email and website. Thanks!
@LennyRachitsky: I came here looking for the same solution. Both 42 and 44 are incorrect links. I'm specifically interested in "42. Running a branding studio". However it shows an incorrect URL (http://run%20a%20branding%20studio%20for%20s/)
The same goes for #44, which @MattLong mentioned here. It also has an incorrect URL
Fixed! Thank you for the heads up.
for complex tasks, there’s nothing like got5-codex-high, IDE from my experience . insanely good!
Curious about this. Is it that Claude Code has a unique ability to do the above, or is it that it the same can be achieved just as well with something like OpenAI’s models, especially as they’ve recently adopted the Agent Skills Convention?
funny to come back to this thread. I fully switched to Claude Code since the holidays break, let's see how this evolve.
Hello @lenny and community,
while I'm a heavy user of claude code and love it, I'm quite surprised to see it recommended for all these non coding tasks. I'm not questioning that it works, it clearly does, but it seems very inefficient and potentially counterproductive.
As we know, a large part of what makes claude code, or any other AI tool, effective is its prompt. Claude code is optimized for coding and carries a lot of tokens and instructions in its system prompt that are hardly relevant to blog writing or image optimization.
It seems to me that for most of the use cases listed here, using claude for desktop would be far more efficient and effective (lower token counts, less context pollution, more usage/hitting window limits, better output). It has all the same benefits - it can run locally, deal with large files, has MCP and connections to tools like linear or github (and I have done most of the things in the uses cases myself with it) so no downsides other than not running in a terminal (which is a big plus for some).
Is there something I misunderstood about the recommendation? has someone concluded that claude code indeed performs better for these use cases than Claude desktop?
Good feedback, Spike! I think much of this is technically true (you can technically achieve many of the same things in Claude), but there's something about the experience of using it as a terminal-based chatbot that feels very different, and makes a lot of things described here easier. I've never imagined doing most of these things in Claude itself, and having it running like this opened up my mind to what Claude can do.
Thanks, Lenny, really enjoyed this one. Love the practical step-by-step guide to open as well.
One small clarification though, since this is aimed at non-technical folks. Where the article states Claude Code “runs locally,” the Claude model itself isn’t actually running on your computer. The model still lives in Anthropic’s cloud. What’s local is the desktop app that can read your files, manage context, and handle all that clever retrieval work.
It’s an important distinction, since otherwise people might think it’s fully offline or entirely private. In reality, only the relevant snippets are sent to the servers while the rest stays on your machine.
Still, totally agree with your main point the local integration layer is what makes it feel so fast and capable.
Really good point, Kanishka! I should have been clearer about that nuance. That that model itself isn't running locally. I'll update the post to reflect this.
This was very inspiring. I have a question about ecosystems, especially since I’m used to OpenAI’s models and API.
Is Claude Code uniquely capable of the above, or can OpenAI’s models, especially with their recent adoption of the Agent Skills Convention, achieve the same?
I want to avoid switching ecosystems unless there’s a genuine advantage to using Claude Code, in which case I’d be happy to.
Mark, great comment. The key is to discern what tasks can be done uniqely by Claud Code and which can do be accomplished using other tools .
Thank you so much, Lenny! This piece really shows the breadth of ways people are weaving Claude Code into their workflows from teachers drafting lesson plans to designers whipping up prototypes. I especially loved the example of marketing teams using it to generate campaigns on the fly. As a non coding PM I have built my own command center around Claude Code + Wispr Flow. Markdown is my single source of truth for everything from PRDs to roadmap decisions and I spin up purpose built agents (what I call Claude Go) to handle specific tasks like requirements checks, surfacing customer insights, suggesting test cases and aligning the roadmap. The speed and depth of insight these agents provide is incredible. I am convinced that tools like Claude Skills and the new ecosystem around it will transform how we work if we embrace them with an open mind. At Why Not Us Labs we are exploring ways to make these powerful tools accessible to everyone and I am always happy to swap notes with anyone curious about how to harness this tech 🚀🕺🚀.
Will Claude code download transcripts from my Apple podcast ?
Try it! Let us know how it goes.
Yup going to try it this week :)