Let Manus write it — I'll own the ingestion
Draft the long-form somewhere else if the tool is better at it. Just never let the tool own where the words land.
By Andrew Pyle
I am happy to let another tool write a first draft. I am not willing to let it decide where the words live.
01Two jobs, not one
Drafting and owning are different jobs
There is a good tool for producing a long, structured first draft from a pile of notes. It is not the same job as publishing. Publishing is where the words enter my system of record, get a schema, get a canonical URL, and become something I can revert, re-date, and re-render on my terms. Confusing those two jobs is how you end up locked into someone else's export format.
So the rule I settled on: draft wherever the drafting is best, then bring the result home through a door I built. The door takes a folder — markdown, a little metadata, some images — and turns it into exactly the same fixture shape every one of my hand-written essays already uses.
02A door I built
The contract is boring on purpose
The ingestion path is deliberately dull. A markdown body becomes structured content the way it always has — headings become sections, paragraphs become paragraphs. Images get copied into the site's own static folder and their links rewritten to my paths, so nothing depends on an external host that could disappear. Metadata maps to the fields I already have. The output is a file identical to what I would have typed by hand.
Because the output is identical, everything downstream is unchanged: the same idempotent upsert, the same revert snapshot, the same prerender and sitemap. The new tool did not get to invent a workflow. It plugged into the one that already works.
The failure I was designing against is the one everyone who has used a hosted content tool knows. You accept its format because it is convenient, and a year later the export has changed, the image URLs point at a bucket you do not control, and moving off it is a migration project instead of a decision. Copying the images into my own static folder and rewriting every link to my own paths is the small, boring tax that buys me the right to walk away at any time and lose nothing.
03Interchangeable by design
The tool is replaceable; the pipeline is not
This is the same instinct behind packaging prompts into reusable skills: keep the durable thing durable and let the disposable thing be disposable.
This is one more instance of directing the work instead of doing every keystroke: the drafter is a convenience I can swap out next month for a better one, with no migration, because it only ever hands me a folder. What it can never own is the schema, the database, and the reversible path from a draft to a live page. That part is mine, and keeping it mine is what lets me say yes to a new tool without flinching.
Related
writing
The moat isn't the data, it's the cleaning
Everyone in a data category has access to the same raw numbers. The thing that's actually hard to copy — the thing worth building a business on — is the unglamorous, endless work of making those numbers trustworthy.
writing
The product isn't the scoreboard. It's the pipeline behind it.
OZ Intelligence looks like a live sports site. The site is the demo. The real product is the data layer underneath — and the data layer is the durable asset.
writing
The scores were never the point
I've been copying down box scores since I was seven. Everyone has the same scores — the data is a commodity — so The Sports Machine is my attempt to give the thing I've loved my whole life the home it deserves.
writing
Someone has to make the data trustworthy
There's a glamorous half of working with data and an unglamorous half. OZ Intelligence is my bet that the unglamorous half — making the data clean and trustworthy — is where the real value hides.
writing
The false comfort of a precise number
I built an atlas of ~3,000 battles to finally see military history laid out in space. The hard part wasn't the maps — it was refusing to pretend I knew things I didn't.