I like writing these notes. I didn’t think I would — thought it would be an annoying, time-consuming chore that takes me away from building. Who needs another blog post? Is anyone going to get value from this? Couldn’t my super intelligent friend named Fable 5 burn the midnight tokens and give me ten variations of ten posts? (Answers: no one, who knows, and absolutely.)
So why am I typing this out? At the most basic level there’s a sense of pride. My parents successfully drilled in the importance of writing from a very young age. I didn’t love it, but it was how things worked. I still remember desperately wanting this new Transformers toy. My mom told me she would buy it if I wrote her a piece describing why I just needed to have it. I can’t imagine it was the greatest thing I ever wrote, but it did the trick. Writing wasn’t just a way to express myself, it was currency.
Writing this is different. It’s not currency and I don’t expect to be gifted any shiny new toy (though if anyone at Anthropic is reading this, I’d love to preview Fable 5.1). But I do think there’s still value in writing to express yourself. If I was using AI, I’d probably describe the topic at a high level, maybe walk through an outline and some specific talking points, and then let Claude cook up a handful of drafts for me to edit and stitch together. Hopefully I’d get something pretty ok (and not just complete slop). The words on the page wouldn’t be mine, but the ideas would be — right? I’m not sure. Easy to second guess yourself in the AI echo chamber.
In a world where I can speed run from idea -> PRD -> working prototype, is there still value in writing things down? I mean — I don’t even pretend to make a PRD anymore (keep in mind I’m only one person, so it is shockingly easy to achieve org-wide alignment). Instead, I go through the process I just outlined in my last post where I talk through everything with Claude until we have something well-scoped to plan and build against. And I feel like I’ve gotten genuinely great results doing this. Maybe I could shave some iteration cycles by typing my prompts out in a more thoughtful way, but I think I’d be just as likely to lose some of the energy and context that I naturally share when dictating (even if it’s messy).
I’m choosing to write these notes here because they are thought pieces, and while LLMs have gotten pretty good at being thought partners (this is how I use Claude much of the time), AI is famously jagged and writing doesn’t seem solved yet. I don’t know why this is. Maybe it’s that ingesting all the writing on the internet leads to messed up ideas of what good looks like. Maybe it actually is good but it’s now too much of a good thing (for the record I do believe this is the case with em dashes — I love them and feel the need to reclaim). Maybe reading this you’re desperately wishing that I’d asked Claude to put these good writing skills to use and take over for me. I’m a believer that writing will be solved, but I’m not sure if I want it to be. LLMs being awkward writers gives me the perfect excuse to actually take a step back and reason through things that matter. By rambling. By writing.
When building things is easier than ever, we need to remain actively involved with deciding what is worth our weekly token limit. That requires thinking — and for me writing forces me to think. Use it or lose it.
Note: I’m also forcing myself to actually write these pieces as a way of accumulating some personalized training data. Something that hasn’t changed since my Zocks days is the importance of being able to show an agent what good looks like, and while they might not all be good, they certainly sound like me. I think building a corpus of great personal data is the #1 most important thing when it comes to making AI work better for you. More on that next week — it deserves its own post.
# Use it or lose it.
> Why I'm still writing these myself when Claude could. A Transformers toy, the death of my PRDs, and writing as thinking.
- written by: Andrew Schroeder
- edited by: Fable 5, GPT-5.6 Sol
- status: published · 2026-08-06
- canonical: https://moltolabs.ai/notes/use-it-or-lose-it/
- raw markdown: https://moltolabs.ai/notes/use-it-or-lose-it.md
---
I like writing these notes. I didn’t think I would — thought it would be an annoying, time-consuming chore that takes me away from building. Who needs another blog post? Is anyone going to get value from this? Couldn’t my super intelligent friend named [Fable 5](https://www.anthropic.com/claude/fable) burn the midnight tokens and give me ten variations of ten posts? (Answers: no one, who knows, and absolutely.)
So why am I typing this out? At the most basic level there’s a sense of pride. My parents successfully drilled in the importance of writing from a very young age. I didn’t love it, but it was how things worked. I still remember desperately wanting this new Transformers toy. My mom told me she would buy it if I wrote her a piece describing why I just needed to have it. I can’t imagine it was the greatest thing I ever wrote, but it did the trick. Writing wasn’t just a way to express myself, it was currency.
Writing this is different. It’s not currency and I don’t expect to be gifted any shiny new toy (though if anyone at Anthropic is reading this, I’d love to preview Fable 5.1). But I do think there’s still value in writing to express yourself. If I was using AI, I’d probably describe the topic at a high level, maybe walk through an outline and some specific talking points, and then let Claude cook up a handful of drafts for me to edit and stitch together. Hopefully I’d get something pretty ok (and not just complete slop). The words on the page wouldn’t be mine, but the ideas would be — right? I’m not sure. Easy to second guess yourself in the AI echo chamber.
In a world where I can speed run from idea -> PRD -> working prototype, is there still value in writing things down? I mean — I don’t even pretend to make a PRD anymore (keep in mind I’m only one person, so it is shockingly easy to achieve org-wide alignment). Instead, I go through the process I just outlined in [my last post](/notes/how-are-we-really-working/) where I talk through everything with Claude until we have something well-scoped to plan and build against. And I feel like I’ve gotten genuinely great results doing this. Maybe I could shave some iteration cycles by typing my prompts out in a more thoughtful way, but I think I’d be just as likely to lose some of the energy and context that I naturally share when dictating (even if it’s messy).
I’m choosing to write these notes here because they are thought pieces, and while LLMs have gotten pretty good at being thought partners (this is how I use Claude much of the time), AI is famously jagged and writing doesn’t seem solved yet. I don’t know why this is. Maybe it’s that ingesting all the writing on the internet leads to messed up ideas of what good looks like. Maybe it actually is good but it’s now too much of a good thing (for the record I do believe this is the case with em dashes — I love them and feel the need to reclaim). Maybe reading this you’re desperately wishing that I’d asked Claude to put these good writing skills to use and take over for me. I’m a believer that writing will be solved, but I’m not sure if I want it to be. LLMs being awkward writers gives me the perfect excuse to actually take a step back and reason through things that matter. By rambling. By writing.
When building things is easier than ever, we need to remain actively involved with deciding what is worth our weekly token limit. That requires thinking — and for me writing forces me to think. Use it or lose it.
Note: I’m also forcing myself to actually write these pieces as a way of accumulating some personalized training data. Something that hasn’t changed since my [Zocks days](/notes/the-ai-slot-machine/) is the importance of being able to show an agent what good looks like, and while they might not all be good, they certainly sound like me. I think building a corpus of great personal data is the #1 most important thing when it comes to making AI work better for you. More on that next week — it deserves its own post.