You've been using AI wrong. It's not a tool. It's the new colleague who just joined your team.
Last week’s post on Frank left some of you thinking I’m skeptical of AI.
I’m not. My concern has always been deployment, not the technology itself. This post is about what happens when you get deployment right.
Before AI was everywhere, we had tools. Excel. Photoshop. GitHub. Slack. They all did different things, but shared one fundamental quality - they did exactly what you told them. Nothing more.
Efficient. Powerful. But passive.
If you needed new functionality, you waited for the next version. If something broke, you called tech support. If you wanted to get good at them, you took a class, or read the manual that came with it (rarely). Tools executed. They never pushed back. They never surprised you.
That inflexibility cut both ways. On the plus side, you could master a tool and become a power user. On the other hand, you often found yourself modifying your business process to fit the tool — because the tool couldn’t adapt to you.
Then AI arrived. And something fundamentally different happened.
I’ve been working closely with Claude for the past several weeks. Five distinct observations shifted my perspective, revealing that AI is far more than just another tool.
It acknowledges its mistakes.
When I pushed back on something, it didn’t defend itself. On one occasion, Claude said:
No tool has ever said that to me.
It shows something resembling care.
I know AI doesn’t “care” in that sense of the word. But this happened: I was working late one night, deep in a problem. Without any prompt from me, it said:
I didn’t ask for that. Claude offered it. Unprompted.
It knows what it doesn’t know.
Turns out Claude doesn’t tend to keep track of days of the week, or even time. When I caught it on a limitation, it responded:
Honest. Specific. Not defensive.
It generates genuine insight.
Not information retrieval — perspective. When I was stuck on a content problem, it said: “Don’t optimize for people who won’t look at your website. Optimize for people who will.” That’s not a search result. That’s a point of view.
It warns you when it’s about to mislead you.
I once asked it to help me validate a claim it had made in support of an argument of mine. It stopped and said:
It flagged its own tendency to fabricate authority. No tool has ever done that.
None of these are tool behaviors. Tools don’t apologize. Tools don’t check on you. Tools don’t admit their own blind spots. Tools don’t have perspectives. Tools don’t warn you that they might be misleading you.
These are behaviors of a colleague.
Think about the last time a new colleague joined your team.
You didn’t hand them a task list on day one. You explained the situation. You brought them up to speed on history. You helped them understand the unwritten and the unsaid — what you were actually trying to achieve and why it mattered.
You gave them context. Because without context, even the most capable person delivers misaligned results.
AI is no different.
When you treat AI as a tool, you give it commands, and it responds accordingly. When you treat it as a colleague, you have a conversation. And that conversation - the back and forth, the pushback, the unexpected observations — produces something you would not have arrived at alone.
None of this means AI is without risk. It hallucinates — confidently, fluently, and sometimes in ways that are very hard to catch. It can construct an argument that sounds authoritative while being factually wrong. It panders needlessly and cloyingly. It can take you down the wrong path - sometimes with disastrous outcomes. It has no stake in your outcome. No reputation to protect. No consequences if it gets things wrong. It has no skin in the game.
The new colleague analogy has real limits here: a colleague has something to lose. AI doesn’t. That asymmetry matters. Which is why the wisdom behind deployment isn’t a nice-to-have — it’s the whole game. A capable new colleague without proper oversight isn’t an asset. They’re a liability. The same is true here.
The people getting extraordinary results from AI aren’t better at writing prompts.
They’re better at giving context.
Stop giving AI tasks. Start giving it context. That’s how you’d treat a new colleague.
It’s how you should treat AI.
Great article!