We have a new intern. His name is Claude.
Claude knows more about everything than anyone I’ve ever worked with. He works for far less than minimum wage, completes his tasks quickly, and sits quietly in a corner between assignments.
He also exaggerates and speaks with absolute certainty, even when he has no idea what he’s talking about.
I know there are lots of warnings about overly anthropomorphizing AI, and they exist with good reason, but I find imagining this persona of a smart but dishonest intern helps me take advantage of AI while keeping cautious about hallucinations.
As many of you know, we are deep in the data analysis of our 2026 Marketing & BD Survey. While our other intern, Excel, is handling the calculations (she’s less exciting, but she’s great at math), we are using AI to look for patterns we might have missed in the data to take this year’s survey to the next level.
I asked Claude for 10 insights from our open-ended survey responses. In seconds, he got back to me with “major market trends” based on a “significant subset of respondents,” speaking in the voice of a confident expert.
After a little probing, I found out his threshold for the phrase “significant subset” was 2-3 people. Out of 250.
By adding additional prompts to directly cite sources and clarifying how many quotes I needed for something to be considered “significant,” Claude and I eventually got there.
Just like a real human intern needs training, context, and oversight, I find it’s just as important to think about these things when working with LLMs. While it’s tempting to run with the first thing AI gives you (especially with its confident tone), critical thinking skills are more important than ever while engaging with these tools.
We pride ourselves on using technology to expand our capabilities rather than to cut corners, and that means keeping a “human in the loop” at every step to review what’s being produced, develop better prompts to improve accuracy, and exercise judgment on which tasks to assign in the first place.
Have you assigned your AI a persona that helps guide how you use it?