I was am (was?) a guest on an upcoming episode of the Vodacom Tech Talk podcast and I was describing my personal AI workflow. I haven’t articulated it to our DM Insiders, so this is that disclosure - also a response to some criticism I have received after publishing analysis of budget documents for SA’s two biggest metros…
Do I use AI in my work? Yes I do. At Daily Maverick we are encouraged to explore ways of using AI to enhance operational efficiency.
Do I use AI to write? Yes, but not the way you would expect.
Let me explain by using my Budget Speech coverage as an example: when you go into budget lockup, journalists must hand in their phones and are not allowed to connect to the internet. So all the interviews/briefings/information sessions must be recorded using only the transcription and audio capture tools on your laptop or you write things down like you’re a time traveller from a different age.
This then showcases the primary step of every story I cover. I go through all the documentation, notes and transcriptions that I have access to and start cataloguing important bits of information, surprising quotes and anomalous data points >> Then I transfer those catalogued findings to what is a master briefing document where I try to sequence the catalogued data in a rudimentary story structure and start cross referencing against my preexisting notes on the same topics (I am a very busy notetaker and keep everything I can backed up in two different apps so I can surface them easily) and my memory.
This is usually where AI comes in. I will put all my notes, audio recordings and the research documents into NotebookLM and break the master briefing into different prompts to double check against the documents in case I missed something or am misinterpreting something - I also ask the AI to surface any other relevant data or quotes that I missed.
Of course, without an internet connection, NotebookLM was a non-starter on budget day, so I rely on conversations with colleagues to identify whether my findings in the budget held up as genuinely insightful and interesting when I say it out loud to other people who are also studying the same documents. Once my research findings passes the water cooler test, then we get to the meat and potatoes.
In most cases I will then open up a blank word document and begin writing my first draft - which is just to get all the information on the page >> Then I refine the draft by adding in the connective narrative tissue and do the makeup so that the reading experience is engaging, informative, entertaining and accurately presents the facts and quotes.
But surely you can use AI to shortcut a lot of the grind work? No, because I can’t write a story when I don’t have a clear understanding of the underlying facts that support the narrative.
Everyone’s craft manifests differently and we all rely on certain structures to be able to execute our tasks with consistent quality. My process helps me meet my personal goals and therefore I cannot hand off the parts that help me arrive at the final product.
That’s why I also take handwritten notes - on my iPad - alongside an audio recording because it helps me remember the vibes of the way the information was delivered.
It’s also why I advocate for technology free classrooms because writing and thinking are interconnected - I believe that you can’t truly understand something unless you’ve written it down.
Has Gemini’s deep research capabilities allowed me to surface data and documents that exist outside of my personal echo chamber? Yes it has. But I find that going down that particular rabbit hole can add hours to my workflow through sheer information overload. So I do boring things like direct searches on trusted databases and then manually download the documents.
Most times I do get sent information from sources or contacts and I prefer to nurture or build those relationships.
And once I’ve submitted an article, I treat my ego by uploading the draft to Claude to analyse it and tell me what a great writer I am.
