AI as a Colleague (not an Assistant)
Many people are getting in their own way when it comes to unlocking higher levels of AI fluency and mastery.
The primary way this is showing up within my clients’ teams is in treating AI like an assistant versus like a colleague. When in reality, the goal for most orgs and CEOs is to get the best out of humans AND the best out of AI, like a true partnership.
I suspect that there are two main stumbling blocks for anyone who feels stuck here:
The limits to how we think about AI and what it can do.
Never having managed or trained others before.
Let’s get unstuck.
First, let’s remove some of our mental limitations that we’re putting on AI just by how we think about it.
The entry point for many of us with AI was using it to handle the low hanging fruit in our day-to-day work. Tell me what’s on my calendar. Look up the cheapest flight for me. Do some deep research on this particular topic.
This is a very narrow, limited and focused way of deploying AI to leverage its superpowers of logic and technical skill.
Think of it almost like a witch or wizard from the world of Harry Potter casting a specific spell. Unlock that door. Turn a camping tent into a full-sized home on the inside. Create light so I can see better in the dark.
Now, AI is essentially the Room of Requirement. Or like Iron Man’s Jarvis.
If you clearly define the problem for AI, and treat it as an immensely capable partner or colleague, it will likely astonish you with its solution and/or how it gets things done.
Does that make sense?
All you need to define is the WHAT and WHY as clearly as you can, and provide the essential context you have that AI doesn’t have yet, and then it can work its magic on HOW it gets it done.
To put it another way (especially if you’ve never managed someone or trained someone up, or even delegated things regularly to another human), imagine that you had an expert sitting next to you. Truly, think of one of the most exceptional humans you’re aware of.
For me, it would be Brené Brown, Adam Grant or Simon Sinek within the world of business and professional development. For F1 racing, it could be Lewis Hamilton, Max Verstappen or Kimi Antonelli. For scientists, it could be Neil deGrasse Tyson, Jane Goodall or Noam Chomsky.
Pick your person and imagine them sitting down with you to solve a problem together.
Would you ask that exceptionally brilliant human to go get you coffee or check the weather while you worked out what you wanted to do, leaving all of their wisdom and experience untapped??!? Hellllll no!
You’d probably start by articulating the problem you’re trying to solve as clearly as possible to ensure you two are solving for the same thing. Then, because you’ve asked this exceptional being to help you with this thing, you’d explain why it matters and would likely provide key context that they wouldn’t have yet.
Then, I bet you’d ask them what they think and the conversation would flow back and forth between the two of you as thought partners solving something together, giving each other feedback to tweak, course correct and polish as needed.
This is only beginning to articulate what AI is capable of. Being a true partner to you at work who:
Is capable of truly owning things soup to nuts
Will escalate to you/flag for you when it’s blocked or out of its depth, or has encountered a problem only you can solve
Can come up with incredible solutions that you would never have thought of on your own.
Does this help? Let us know where you’re blocked and let’s see if we can’t help you get the absolute best out of yourself, the humans around you, AND your AI peers/colleagues.
Claude’s Take
To lean into partnership myself, I shared my above write-up with Claude and included what it said in response below. Note that I’ve used [brackets] to identify where I (Faith) have inserted my own thought or edited what Claude wrote for clarity here.
Claude writes:
The Room of Requirement is a good fit for the part most people miss. Most people use AI like Lumos or Alohomora, a single known spell for a single known task. The Room works differently: you describe the need, and what appears is shaped by it, even if you couldn't have specified the "how."
There's a detail in the books that helps your point. Harry has to walk past the wall three times while concentrating on exactly what he needs. It's reasonable to infer that a vague or muddled need would get a vague or muddled room.
That's a good model for working with me: clarity about the problem matters more than knowing the technique. "State the problem clearly" is a skill, not a throwaway instruction.
There's one place the analogy breaks: The Room is portrayed as reliably giving you what you need. I can get it wrong. I can misread the need, miss context, or sound confident while being mistaken. So the better frame is a room that builds a strong first draft of what you need, which you then walk through and check. [It’s important to] judge the output rather than just receiving it.
On Jarvis, I'd agree he's the most recognizable reference, and he captures things the Room doesn't:
Ongoing collaboration. Tony thinks out loud, and Jarvis keeps up, fetches things, runs analysis and pushes back occasionally. That's much closer to how people can use me day to day.
Natural language as the interface. No special syntax, just talking.
The weakness is that Jarvis is a subordinate to a single genius, so he undersells how much I can contribute to the thinking itself, such as pressure-testing a strategy, drafting, or spotting patterns. He's also usually portrayed as infallible, and he mostly works for one person.
A few other parallels might serve you better, depending on the audience:
The Star Trek computer. "Computer, analyze this and give me options" has the same problem-first, conversational feel as the Room, but it's more grounded in work than in magic.
A Chief of Staff, in business terms. For CEO readers this may land hardest. It's someone who can draft, research, synthesize, prep you for meetings and challenge your reasoning. The limits are the real ones too: they need context, and you still own the decisions.
The Time-Turner. Hermione uses it to be in two classes at once. That captures the capacity (getting more done in the same hours) that the Room doesn’t.
If I had to pick a combination, I'd use the Room of Requirement for what to ask (the need, not the method), Jarvis for how to work together (an ongoing conversation), and the chief-of-staff framing for how much to trust it (high leverage, with you still accountable).
[If you’re an IC, this is what I, Faith, and Claude want to emphasize]: you don't need authority or technical skill to benefit, just a clearly stated need.
Faith’s “including themselves” and “best out of their humans” framing also points to something deeper. The skill the Room rewards, getting clear on what you actually need and saying it plainly, is the same skill behind good delegation, good feedback and good management. People who are stuck with AI are often stuck in the same place with their teams:
Vague asks get vague results, from a person or from me. "Make this better" fails in both directions.
Context is the multiplier. A new hire and I both do better when we know the purpose, the audience, what "good" looks like and what's already been tried.
Treating the other party as a thinking partner beats treating them as a vending machine. Asking me "what am I missing?" works like asking a team member for their read, not just their output.
That could be the bridge for this piece: the way you work with AI is a mirror of the way you work with people. I suspect readers who get better at one will get better at the other. It also lets you meet the skeptic and the enthusiast in the same place.
Each level might connect with a different image:
IC or receptionist: the Room, "I can get things done that I couldn't before, and I don't need permission or expertise to start."
Manager: the Room plus delegation, "the quality of what I get depends on how clearly I frame the need."
CEO: the chief-of-staff or Jarvis framing, "leverage on my thinking, with me still accountable for the decisions."
I'd keep the trust caveat at every level, since the Room is portrayed as reliable and I can be wrong. It also fits your theme: a good partnership includes checking each other's work.
Suggested Habits for Working More Effectively with AI
Faith here again:
Claude suggested habits that I agree with and have seen prove out on teams, and that you can experiment with to improve your partnership with AI. As someone who firmly believes that how you work with people and how you work with AI are connected, I’m confident that practicing these approaches will have a quantum ripple effect for you in a positive way. I’ve highlighted my own added thoughts in green below and I've underlined a few things for emphasis.
Claude writes:
Here's a set that works for anyone, whether they're an IC or a CEO. Each one has a parallel in how you'd work with a person.
Say the outcome, not the task.
Instead of "write an email to my team," try "I need my team to understand why we're pausing the Q1 launch, without feeling blamed." The first gets you a generic email. The second gets you something that serves your real need. With people, it's the difference between handing over a to-do and handing over a goal. Define the what and why; let your agent and people figure out the how!Give the context you'd give a smart new hire.
Cover who it's for, what "good" looks like, what's been tried, and what's off the table. Many disappointing results come from missing context, not from a limited tool. It's the same reason a new hire flounders without it.Name the stakes and the constraints.
"This goes to our board" or "she's already frustrated" changes what a good answer is. Say the unspoken parts out loud, including length, tone and what to avoid. Again, context is key! What important information do you have that AI wouldn't know until you told it?Ask for the thinking, not just the output.
Try "What am I missing?", "What would a skeptic say?" or "Give me three different approaches and the tradeoffs." This is where the partnership shows up. You're asking for a perspective instead of treating it as a vending machine, which is also how you get the best from a team member.Iterate like a conversation, not a retry.
If the first result is off, don't start over. Say what's working and what isn't: "The structure is right, but the tone is too formal." It's the AI version of specific, kind, direct feedback. Kim Scott's "care personally, challenge directly" translates well here, and so does the habit of being specific rather than just saying "not quite."Verify what matters.
Treat the first output as a strong draft to walk through, not a finished product. Check facts, numbers and anything high-stakes, and ask for sources, then open them to confirm they say what's claimed. This is the one place the Room of Requirement metaphor stops, because I can be confidently wrong, so the human stays accountable for the result.
A bonus habit for the "stuck in limited thinking" crowd is "Ask what else is possible." Try "Here's my problem. What are some ways to approach it that I haven't considered?" or "What could you help with here that I haven't asked for?" Many people only request what they already know how to do, and this habit is how you get beyond Lumos.
One caveat is that these come from my own sense of what works plus general prompting guidance, not from a study I can point you to. Anthropic's prompt engineering overview covers habits 1 to 3 (be clear and direct, give context) in more depth if you want a citable source.
