I count Lead Dev London 2016 as my “breakout” conference. I gave a talk about The Seven Righteous Fights, which I’ve covered in more depth at the #7fights tag. So imagine my delight when I was invited back for Lead Dev’s 10th anniversary bash!
I talked about technical writing and AI. The video is here and the slides are here. You’ll need a LeadDev account to view the video, but it’s free, and honestly, you’ll benefit from a lot of amazing content.
The main thing I wanted to convey in this talk is that the hard part of technical writing is not figuring out what order to put words in. Indeed, an LLM can do a medium job of that given a sound prompt. The hard part is thinking through the context that makes it useful to the people reading it.
I think that it’s easy and buzzy to think about AGI or generative tools. That’s not where I see the real value of machine learning. I think it’s going to be in specific tools that have a narrow focus and can do as well or better than humans at a small set of tasks, precisely because they can sift so much data.
What that means for technical writing is that there are a lot of places you can use “AI” to write documents for procedures and topics that are predictable, well-formed, semantically reliable.
- APIs? Yes, absolutely.
- Procedures that have stable and mature test suites? Sounds great.
- Release notes in a system with trustworthy and well-defined bug tracking? Probably.
- Tracking the programming decisions made that got everything the way it was? No, but humans don’t do that either, so whatever.
- Conceptual explanations that cross domain boundaries? No. Please don’t try.
I then took some time in the talk to go through tools that are (as of mid-June 2025) useful for parts of the writing process.
Mostly, I want you to remember that generative AI is risky, because it doesn’t know what is true, but that other kinds of machine-assisted writing can be good and useful if you don’t imagine they’re magical.