On Solowriting
In 2026 I'm reinstating my blog and hoping to share more of my thoughts and own work about research, AI, productivity, parenting and other top of mind issues. My intention is for this to be a space for writing that is 100% LLM free.
I'm not sure if there is an existing word that spells out the distinction between writing that is assisted vs completely unassisted by LLMs. I feel like we should have a name for this, just like when automation or mass production has entered other fields, the result was new language to mark that something is 100% hand-made—like a hand-crafted scarf or bespoke suits. For now, I'm calling this solowriting:
Solowriting is the process of fully producing a piece of writing without using language models or generative AI in general in the process of ideation, structuring, writing and proof-reading. It stands as a direct antithesis to writing together with AI as the "co-pilot" (and eventually the pilot itself).
In this (solowritten) blog post I want to share more of the reasons behind this choice.
My most immediate reason to practice solowriting is the realization of how much I'm relying on AI for writing-related tasks. Writing is a big part of my day to day job—research logs, project updates, papers, random ideas—and I'm relying to AI to help with all of these tasks (I've even just completed an "experiment" where I tried to get Gemini to write an entire paper for me—more on how that panned out in a future post). While helpful in general, I'm already feeling the "slippery slope" nature of my reliance on AI, where I'm increasingly tempted to run even shorthand or simple texts by it, like an email to a senior colleague or even a tweet. This is starting to have a flavor of needing the LLM's confirmation on everything written, which is a quick and sure way to decrease my overall agency and autonomy. So flexing my ability to solowrite feels like an important pushback against this process. It feels a bit like taking a drive without a GPS based navigation system. It's something we all are in principle capable of doing (and did constantly before smartphones entered our lives), but can now feel a bit intimidating.
One natural question to ask, I think, is why the great concern—if I'm fine offloading coding tasks to an LLM, why not writing too? For me, personally, the difference feels fundamental. Coding is mostly a means to an end for me—I want to get something done, so I need to write code to do it. If there is now a new way to get the same thing done just as well without me writing the code myself—great. But with the type of writing tasks I work on, writing is thinking, making it an end in and of itself. There is no neat boundary between "thinking about the text" and "writing the text"—the two are inherently intertwined, and working my way through the writing process is a tedious process that hones in my arguments and fleshes out my thinking. So giving up on substantial parts of the writing process is essentially giving up on the thinking too, and it's something I'm by no means ready to do (and I also think the current state of LLMs are still quite poor at).
In general, the proliferation of AI generated text everywhere we look has dramatically increased the value of certifiably human-written texts. I personally can't help but cringe when someone posts text in their own voice and under their own name, but it has all the telltale signs of LLM writing, such as excessive use of metaphors, predictable structure and low information density (see Shreya Shankar's wonderful recent post on this topic). So I almost feel some kind of (perhaps silly) moral obligation to stay somewhat away from this madness myself and force more "friction" into my life.
Reflections
Wait, you truly didn't use an LLM here? That's right, I wrote everything in a single Google doc. I then asked Claude to take the text and update my website with it. It told me it applied some formatting ("proper em dashes for readability") and concluded by saying that "The irony of using an LLM to format a post about solowriting is not lost on me!", which is such a nice and humorous touch, and honestly one of the reasons why I love Claude.
So, how did it feel to solowrite this blog post? In one word—hard! There were quite a few times that I felt the temptation to move forward quicker by having the AI help me out. "Ok I wrote down some of the main ideas, let's have it complete the entire post for me". In a sense I'm quite happy that I found it this difficult, because it tells me that the effort is indeed worthwhile—kind of like going to the gym after a long break, being surprised at how bad you suck, only to realize that you should really keep coming back. So I'm definitely looking forward to flexing my solowriting skills more in 2026.