|

Why you should stop treating AI like a person

I was talking with a friend recently about how easy it is to treat AI like a person.

We call it an assistant, a teammate, even a brain. If you expect it to behave like a human colleague, though, it’s going to keep letting you down.

I think a better mental model is a very particular kind of alien intelligence.

How the alien thinks

  • It’s read almost everything. It’s taken in most of the internet, knows every programming language and handles technical jargon in any field without blinking.
  • It reads a whole document at once. People read word by word and line by line. AI takes in the entire document in one pass.
  • Its memory is short. Its working memory only lasts for the session. Close it and it’s forgotten you were ever there.
  • It’s a master of language and patterns, and it can’t read your mind. It won’t know what you want unless you spell out the goal and the limits.

Once you treat it as a very fast engine for working with language, instead of a human assistant who thinks the way you do, your expectations start to match what it can do. In practice that means spelling out the context and the limits every time.

Where AI struggles and where it does well

Seeing AI as an alien processor makes it easier to understand why some tasks flop and others feel like magic.

It struggles with:

  • Unspoken context. It can’t read office politics or the unwritten rules everyone on your team just knows.
  • Following through over time. It won’t remember to follow up next month unless someone builds that reminder into the system.
  • The human weight of a relationship. It doesn’t feel empathy or understand what a donor or client relationship means to your organization.

It’s very good at:

  • Reading a lot at once. It can take in 50 emails or a 100-page grant report and pull out the key numbers.
  • Turning one format into another. It can turn a raw meeting transcript into clean action items or structured data.
  • First drafts. Templates, outlines and summaries come back in seconds.

So hand the heavy language work to the machine, and let your people spend their energy on relationships and decisions.

How to start: a read-only inbox agent

A lawyer who sits on a nonprofit board asked me recently where she should start if she wanted to see what AI could actually do in her daily work.

My advice was to skip the prompt courses and hand her email inbox to a read-only agent. The AI doesn’t send or reply to anything, ever. That boundary is what makes it safe to try.

  1. Set it up as read-only. The agent can read messages as they come in. It can’t send, reply, delete or move anything. A tool like Claude Cowork or a custom workspace agent can work this way.
  2. Have it sort by priority. Ask it to group new messages so the urgent ones sit at the top.
  3. Ask for a short summary. Two sentences on what actually happened in each thread is plenty.
  4. Have it pull out the decisions. Give it two headings to fill in: “Decisions you need to make” and “Next steps.”
  5. Check its work for the first few days. Compare the digest with your inbox and correct it when it gets something wrong.

Within 48 hours, you stop drowning in 200 unread threads. Instead of spending two hours reading every message in the morning, you read a structured list of decisions in about five minutes.

That one low-risk exercise shows a leader more about what everyday AI can do than a dozen webinars. If you don’t know where to start, start with the work that tires you out the most.

For leaders: set expectations that fit the tool

A lot of AI disappointment comes from expecting it to act like a person. Your team will get more out of it once everyone treats it as a fast, literal reader that only knows what it’s told.

Questions to bring to your team:

  • Which tasks are we giving AI that depend on context nobody’s written down?
  • Where could it read a lot at once and save someone hours?
  • What would a new human hire just know that the AI has to be told?
  • Where is our team’s mental load highest right now?

Designing around what AI is actually good at is how I’m approaching Cosly, the AI staff team I’m building for small nonprofits.

Similar Posts