LLM-assisted writing workflow

Notes on "Writing in the Age of LLMs" by Shreya Shankar: https://www.sh-reya.com/blog/ai-writing/. Her approach to using LLMs closely mirrored my own, but she articulated it very well.

The goal when using LLM assistance is to keep the writing momentum going. Whenever you hit a bottleneck, use the LLM to regain momentum.

Process:

• narrate the high-level idea to the LLM, even if your thoughts are disorganized. Ask it to make a detailed outline.
• iterate on the outline. Don't move forward until it is solid
• manually write each paragraph for the outline, even if it sucks. Don't get too bogged down with editing.
  • if you get stuck with a particular sentence or paragraph, ask the LLM to finish it
• read, critique, and rewrite here
• use scoped rewrites using specific instructions, not vague "polish this".

The first is to put the subject and verb close together, at the beginning of the sentence. The second pattern I use is SWBST: Somebody Wanted But So Then. It’s a basic storytelling structure—often taught in early writing education, but surprisingly effective in technical contexts because it helps convey motivation, conflict, and resolution in a compact form. The “Somebody” is the actor, “Wanted” states the goal, “But” introduces the obstacle, “So” explains the response, and “Then” describes the outcome. In technical writing, this structure makes it easier to show how a decision was made or how a system evolved in response to a problem

Don't delegate what to say, how to frame it, or what to emphasize. LLMs suck at this.

The post also has a whole section describing LLM slop:

• Overusing demonstrative pronouns like this, that, these, those without a clear noun. For example, what is "this" in the sentence “This creates friction in production.”?
• Not using the right subject and predicates in a sentence, resulting in unclear writing.
• Garden variety vagueness, low information density, hallucinations.