Field notes
Not everything needs a model
AI is worth a lot in some places and a quiet liability in others. The line is whether it’s doing something no human could, or just standing in for one who was doing fine. We start with people and bring in the machine second.
Use AI where it makes a real difference. When a model just takes over a job a person was doing well, it quietly becomes a liability, so we lead with people and use the machine where it genuinely helps.
On this page
I don't think you should write your emails with a model.
The model can write it, and the email will come out fine. Fine is the trouble. When everyone runs their thinking through the same handful of models, everyone ends up in the same competent, forgettable middle. The email that gets read is the one that sounds like a person sat down and had a thought, which is the bit you can't hand to a model and still keep.
That is roughly the argument. A model earns its keep when it makes a real difference. It turns into a quiet liability the moment it's only standing in for a human who was already doing the job well.
Use it where it earns its place
Some jobs only a model can do well: cleaning a million records, reading a thousand support tickets to find the three that matter, watching your data around the clock for the anomaly a person would never catch in time. That's a model changing what's possible, at a scale no human could touch. Use it there, and use it hard. That's the case for building AI into the revenue engine instead of bolting it on.
Don't use it because everyone else is
Somewhere in the last two years, "we use AI" turned into something companies say for the sake of saying it. Teams are bolting models onto work that was already working, because sitting it out started to feel like falling behind. Cargo-cult adoption, basically - you go through the motions and skip the part where it helps you. And there's a cost. Every task you hand to a model that didn't need one is a bit of judgment you've stopped exercising.
Human-first is not anti-AI. It's the opposite.
The teams that win with AI don't use the most of it. They keep people on the work that needs judgment and give the machine what it's genuinely good at, and they can tell the two apart. Put people first in a business that runs on AI and the work gets better. Flip it - machine first, human as backup - and you're left with speed and not much underneath it.
Whether AI can do a given thing is rarely the interesting question, because it almost always can. What matters is whether it should, here, for this particular work - and that is a call a person has to make. It's the whole basis of how we use AI at all.