Six questions about AI from people who don't write code
These come up almost every week. Collecting them here so I can stop writing the same answers.
These come up almost every week. Client meetings, family dinners, conversations at the gym. I’m collecting them here so I can stop rewriting the same answers.
Should we replace some of our staff with AI?
Probably not the way you’re thinking. The jobs people imagine AI taking over are usually the ones with the most context, judgment, and relationship work. Those are also the ones AI is worst at right now. What you can do is take specific tasks inside a role (not the whole role) and automate them. The person still works there. They do more, with less of the tedious part. Whether you eventually need to hire fewer people is a separate conversation, and one most leaders aren’t actually ready to have.
Is this going to be like crypto, where most of it disappears?
The hype will cool. The technology won’t disappear. The difference is that AI is already useful to a lot of people for a lot of things. Crypto was useful to a much smaller group. When the speculative money leaves, what was actually useful stays.
Can AI just do my email?
Some of it. It can draft replies, summarize threads, surface what matters. It cannot be trusted to send things on your behalf without you reading them. Not because the writing is bad, but because email is high-stakes and the failure mode is embarrassing. The current sweet spot is “AI prepares, human approves.” If a tool is offering you more than that without a careful review step, be careful.
Why are AI tools so confident when they’re wrong?
This one is a real problem. Models are trained to produce text that sounds plausible, not text that accurately represents what they know. When they don’t know, they don’t say so. They write something that fits the shape of an answer. Better systems are getting built around this (asking models to express uncertainty, checking answers against sources) but the underlying behavior isn’t changing soon.
My team is afraid of data leaks. Should they be?
Depends on the tool and the data. Cloud AI services see whatever you send them. Most consumer-facing ones reserve the right to train on it unless you turn that off. Enterprise versions handle this much better. The practical answer is usually: pick approved tools, train people on what to put in and what not to, and move on. Banning AI doesn’t make people stop using it. It just makes them use the worst version on their personal phones.
How do I tell if a vendor’s “AI” is real?
Ask what they were doing six months ago. If they had a working product that solved a real problem and AI is now making it better, the AI is probably real. If they didn’t exist six months ago and the entire pitch is AI, be skeptical. Real applications usually need real domain expertise, real data, and a real workflow to plug into. None of that gets built in a weekend.