The "messy middle" is AI's sweet spot + my writing story
There are essentially unprecedented budgets and hype around AI right now, and naturally leaders are getting extremely ambitious about what AI can do for their organizations: replace entire jobs, intelligent assistants, augmenting intelligence, phrases like that. This might happen eventually? But right now it’s just biting off way more than the actual AI models we have today can chew.
IMO the best use for AI models today is to automate repeatable, manual tasks that you do in your day to day and don’t require external cooperation.
Personally, (as a writer) I was extremely skeptical of GenAI models when they first came out, and resisted using them for years, really until the beginning of 2025. Everyone was prompting models to generate entire posts, which felt very wrong to me both practically (these posts sucked) and spiritually (they sucked out your soul).
But earlier this year I started to try them for one small specific thing, which was putting together scaffolding for a post before I wrote it, essentially eliminating the “blank page” problem (and writers reading this will understand). With a combination of a detailed style guide and good prompting – and then really getting in and writing – I’m able to get what’s in my brain onto paper faster, while still giving people personal, authentic writing that is actually from me.
The point is, I didn’t try to automate all writing for Technically (yikes), or create some new interface where you just talked to a Technically model. I took one specific thing that was tedious and fairly automatable, spent a few weeks/months tinkering with it, and got it to a tractable, good spot.
The same is true in finance. I spoke to CJ Gustafson who runs the excellent Mostly Metrics newsletter. When it comes to AI for finance teams and CFOs, he thinks there’s this "messy middle" where AI is perfect. To understand the middle, you have to understand the extremes where AI is a terrible fit:
- High risk, low volume tasks: Think about transferring a large sum of money. This happens infrequently, but the consequences of getting it wrong are catastrophic. Letting an AI handle this is dumb.
- Low risk, low effort tasks: On the other end, if a task takes you five minutes, you probably don't need to spend an hour setting up an AI automation for it. Software engineers fall for this all the time…
The messy middle is everything between those two poles. It’s the work that takes up 30-90 minutes of your day, happens regularly, and is mind-numbingly manual. Think: creating the first draft of a weekly report, summarizing a long meeting transcript, or categorizing a hundred pieces of user feedback. This is exactly what I use AI for today.
