The posts you are reading this week were all written on Thursday 8 October. Normally this project produces the following week’s content on a Friday. On the Friday I was travelling to Phnom Penh, so the whole week moved forward a day.
That sounds like a small change. It turned out to be a useful audit, because doing everything a day early showed which parts of the routine need me and which do not.
How a normal week works
The routine has a fixed shape. An AI research agent reads eight sources of AI news every morning and writes a digest. Through the week, candidate stories go into a database with notes on the evidence and the angle. On Friday, five are chosen for the following Monday to Friday.
Each one becomes a blog post, a Facebook version and a LinkedIn version, plus a picture card for each. The social posts publish themselves at 3am Cambodia time, once scheduled. The blog goes up later the same morning through the website.
Much of that is now done with an AI assistant working through the steps alongside me.
One design choice matters for what follows. The research agent writes its report and stops. It is not allowed to add stories to the database itself. A separate step does that, with someone reading the report first, because a story logged with the wrong detail tends to stay wrong for weeks.
So the question that week was simple: if I am not here on Friday, what still happens?
Pile one: what ran without me
Most of it, once I wrote it down.
The research agent kept running. It does not know what day production happens, and it does not need to. The posts already scheduled for that week kept going out on time, because the scheduling tool does not care where I am.
Most of the production itself also ran with very little from me. Drafting the five blog posts and cutting them down for each platform, checking each source, writing the database records, rendering the cards and running the house style checks all happened in one working session.
Two moments from that session are worth describing, because they are the part people tend to assume needs a person.
The news report behind Monday’s post, about Singapore’s financial regulator, did not link to the regulator’s own release. Before any copy was written, the release was found and read on the regulator’s website, and the post is built on that rather than the report. And the note behind Thursday’s post said that more than 100,000 Wipro staff “were trained first”. The Reuters story it came from says they “are receiving” training. That was corrected before drafting too.
Checking against the source does not need me. It needs a rule that says always do it, and a record of what was checked.
The house style checks are the same. Every post is searched for the words and punctuation we do not use, and every Facebook version is counted against a hard limit of 600 words. A script does that more reliably than I would at the end of a long day.
Pile two: what needed my judgement
Three things, and all three were choices rather than tasks.
Choosing the five. I was given a shortlist with reasons: which stories had the strongest evidence, which pillars were behind on the month, which subjects we had covered too often lately. It also flagged that two of the previous week’s posts had been about who checks AI’s work, so a third that week would be one too many. The shortlist was good. I still had to pick, and one of the five, this post, was my choice from three options rather than something on the list.
Choosing the pictures. Every card uses a stock photograph. Those cannot be fetched automatically from the library I use, so I pick each one by eye and export them by hand. That is partly a technical limit. It is also a matter of taste.
Approving the copy. Nothing goes out in my name until I have read it.
Pile three: what needed someone to confirm it
This is the pile that is easiest to forget, and the one moving a day exposed.
Every Friday, before anything else, I check what actually went out the previous week against what was planned. Nothing does that automatically. A scheduled post that fails tends to fail quietly, and the only way to know is to look.
Producing a day early meant that check could not cover the previous week’s Friday post, because on Thursday it had not run yet. Friday’s research digest also arrived after the week’s choices were made, so it rolls into next week instead. Neither is a problem. Both are the kind of thing that becomes a problem if nobody notices they moved.
So the handover was a short written list rather than a feeling that everything was done: which checks were brought forward, which were left for the following week, and which day each one would now happen. It sits in the week’s folder, where whoever picks the work up next can find it.
What I would suggest
If you are going to be away, try sorting your own routine the same way before you go.
What runs without you. Usually more than you think. Write it down anyway, so whoever is covering knows not to touch it.
What needs your judgement. Choosing, approving, deciding what good looks like. These are worth doing early rather than delegating in a rush.
What needs confirming. The checks that something actually happened. These are the ones to hand to a named person, with a date, because they are the ones that fail silently.
What the day showed was that the second and third piles were small. The work did not need me. The choices did, and so did the checking afterwards.