Skip to content

AI in practice

I've used AI tooling seriously for five years. Not experimenting — working. The distinction matters.

What changed

For most of my career, the bottleneck in documentation was time. I knew what needed to be written and had the structure in my head; producing the text at the quality I wanted took longer than projects allowed. So documentation got cut, compressed, or delegated to someone who didn't understand the system well enough to write it accurately.

AI removed that bottleneck. Not by writing for me — by removing the friction between thinking and text. I provide the architecture, the source material, the editorial judgment. AI provides velocity.

The concrete result: the libdrone documentation stack — specification, variables file, SDK, build guides — produced by one person, evenings and weekends. How it was done is described in AI-assisted documentation.

How I actually use it

Architecture first. I design the document structure before writing a word: the sections, what each answers, the single source of truth for each value. AI doesn't design this. I do.

Source material in, structured text out. I feed in technical facts — measurements, component specs, process steps — and get back prose, which I then edit. The editing is significant. AI text without editorial judgment is recognisable and unconvincing.

Consistency checking. A forty-document stack with hundreds of cross-references is an editorial problem. AI is very good at finding contradictions and flagging a value that appears in two places with two different numbers.

The variables file discipline. Every numeric value lives in one file; every document references the name, not the value. Not an AI technique — an editorial one, brought from the magazine world. AI just makes it enforceable at scale.

What it does not do

AI doesn't know when something is wrong. It produces plausible text, and plausible is not accurate. Every factual claim I publish is verified against a primary source — the hardware, the measurement, the datasheet.

AI doesn't have taste. It can't tell that a technically correct sentence will confuse a reader arriving without context. That judgment is editorial, and it's what I bring.

Where I stand

AI doesn't replace editorial judgment, domain understanding, or the ability to design documentation that serves its reader. What it replaces is the excuse that good documentation takes too long. It doesn't, anymore — which leaves designing it properly as the only remaining hard part.