Treating Your AI Like a Colleague (Not a Tool)

Hi all,

I've been thinking a lot lately about the differences between AI agents and humans.

In my keynotes, I share some emerging research that I find genuinely fascinating: when researchers studied how people interact with AI, they found something unexpected. We don't interact with these tools the way we interact with calculators or spreadsheets or search engines. We interact with them the way we collaborate with human colleagues.

We negotiate. We push back. We get frustrated. We apologize (yes, I've apologized to Claude more than once). The mental model is just fundamentally different.

And if that's true, then there are some very human attributes we're going to have to give our agents if we want them to be effective collaborators: memory, personality, and culture.

I've been learning this the hard way over the past two months.

Memory

I started a side project to build an email auto-responder. The premise was simple: I spend hours every day responding to emails, and there has to be a better way.

My early attempts were okay. The drafts were grammatically clean and reasonably on-topic. But they had a fundamental flaw — the AI had no history with anyone in my inbox. So it would respond to a brand-new prospect the same way it responded to a client I'd worked with for five years. Friendly, helpful, and generic.

So I decided to build out a massive vector database — which I call House Party — that combines all my historical communication across every channel I use: Gmail, Slack, iMessage, WhatsApp, Zoom, Teams, Telegram, LinkedIn. If I've ever talked to you, you're in there somewhere.

I vibe-coded my way into the system using Claude as a very skilled data engineer, and I'm thrilled with the result! The difference in tone is striking. Here's a side-by-side of a message thanking my friends at UT for the opportunity to teach during last month's AI Masterclass:

Same task. Same model. Wildly different output once the AI actually knew who it was writing to.

Personality and Culture

Here's the thing I learned along the way: the AI was an exceptional data engineer and a terrible colleague.

It would receive instructions to make changes to the database and eagerly run off to do the job — while blowing away massive amounts of data in the process. I found myself yelling at my computer more than once: "Claude, you just wasted $500, why did you do that?!?"

It was like working with an intern who was technically brilliant but had zero professional judgment. Smart enough to do the work, clueless about how to actually work with me.

Fortunately, Claude has something called a Claude.md file that lets you save details about who you are, how you like to work, things the agent should always do, and things the agent should never, ever do.

Over time, I developed a set of guidelines that I call House Rules. They cover everything from "always confirm before deleting anything" to "never run a script that touches production data without explicit approval" to "if you're not sure, ask."

A few weeks ago, I asked Claude to do something potentially destructive to the database. It paused, came back, and said: "I don't think this is a good idea." I felt like a proud mentor.

The Bigger Point

As we start to see agents more as colleagues and less as tools, we're going to have to think about what — beyond raw skills — makes someone a valuable, trusted collaborator.

It's memory. It's personality. It's understanding the culture and context of the work. And yes, eventually, it's even giving them names (mine is named House Party for a reason).

I'd love to hear about your experiences collaborating with agents — what works, what doesn't, and how you've defined the rules of engagement for your own personal or work projects. Hit reply and let me know.

Best,

Dr. Michael "House" Housman

P.S. Full disclosure: I'm still personally reviewing and tweaking every draft that House Party generates before it goes out. But we're getting closer to the day when you'll be emailing my agent and won't even know it.