Jerome BoyerSimple Interface Connections · Connecting your business to AI, simply.

← All posts

How to Talk to Your AI Colleague

The Rising AI Sea, part 5: from giving orders to giving goals

The AI "brain" changes fast

The AI I work with today is not the one I worked with six months ago. It reasons better, follows instructions better, remembers more of the conversation, reads pictures and documents, and can now act on its own: send a draft, update a file, check a website. Even the same assistant can behave a little differently from one month to the next. So the way we talk to it has to keep up.

From giving orders to giving goals

In the early days, when AI was a hammer or a toolbox (buckets 1 and 2 in my picture, summarized at the end), you gave the AI instructions. Step by step, very precise, almost like programming. You picked a tool and told it exactly what to do.

With an assistant, bucket 3, that changes. Today's assistants understand plain language and what you mean, not just what you type. Too many instructions can even get in the way. You now talk about GOALS, CONSTRAINTS and strategy, the way you would with a good colleague: here is what I want to achieve, here is what matters, here is what to avoid. Then you look at the result and give feedback.

With a chatbot, I worked back and forth in a chat window, guiding it one step at a time. With an assistant, I give it the goal and it does the steps itself: it writes, files, checks and reports back, like a good colleague.

Give the goal, not the detail. In 2024, I was programming a complex application and gave instructions like: "I need the title bar to look like this for this window." The results were NOT pleasing. Jump to 2026. My instruction: "I need a pleasing user interface to describe, simply, in an organized way, the results from these tables." The result was night and day. In seconds, the programming assistant made the visuals: clear, obvious, pleasant to the eyes and focused on the results from these tables. Everything fell in the right place. Same person, same kind of task. What changed was the assistant, and how I talked to it.

I'm not the only one who noticed. Ethan Mollick, a Wharton professor who writes about AI, says that in 2024 an AI could handle a four-minute task, and by 2026 it could handle 17 hours of human work. He now gives it a few sentences of goals and says "make it better," because, in his words, "prompting has become much less important as AIs get better." (30 Months on the Exponential)

But don't chase every new trick

Here is the good news: the basics don't change. Five habits work with every new version:

  • Be clear about the goal, and make it wide. Not "change this paragraph in my email," but "I want this customer to say yes."
  • Give context and limits: your audience, your niche, your budget.
  • Divide to conquer, but don't overdo it. Too many small pieces and you lose the big picture.
  • Check what comes back. You are in charge, all the time.
  • Know which bucket you are in: 1, 2, 3 or 4?

You don't need to relearn everything with every release. You just need to stay aware of what your assistant does well today, and what it still gets wrong.

That is where my assistants and I can help: we keep up with the changes, so you can keep running your business with the best assistant for the job.

Try it this week: pick one task you'd normally explain step by step. Give your assistant only the goal and one limit, then compare the result.

Next: "The spirit of the code", on why AI still needs the person who knows why something was built the way it was.

The four buckets, in short

To make sense of how fast AI is changing, I developed a simple picture: four buckets, from AI as a hammer to AI as a partner. Ethan Mollick sees the same shift from chatbots to agents.

  1. A hammer: one tool, and everything looks like a nail.
  2. A toolbox: many tools, but we see only a few of them.
  3. An assistant: it understands ideas and acts on them. You give goals, it does the work, you check.
  4. A partner in creating NEW things: step by step, like Edison testing thousands of filaments, but with AI doing the testing in seconds.

More: Staying afloat as AI rises · What If Edison Had an AI Assistant? · Ethan Mollick, 30 Months on the Exponential

Start with your story.

Thirty minutes, no cost, no obligation. In English or French. British Columbia, working across Canada.

Book a free 30-minute conversation