Copilot, not Autopilot: Three weeks of learning to trust (and argue with) my AI assistant

A few weeks ago, I started a new job managing a portfolio of pharmaceutical projects , orchestrating the tech transfer process of products from CMO to CMO, and market launch. Anyone who has worked in pharma knows this is a world of handovers, documentation, and enough acronyms to make your head spin. But that's not what this article is about.

During onboarding, I sat down with the IT guy for the standard "here's how things work here" chat, and he said something that caught me off guard: "You can use Copilot as much as you want. It's internal, nothing you type is used to train anything. You can upload confidential data, ask whatever you want, be as explicit as you need to be."

I paused.. In previous companies I worked for, AI tools came with a leash. You could use them, sure, but with a healthy dose of suspicion. So hearing "go ahead, be fully transparent with it" felt like a genuine game changer. And it made me realize something: reading about AI capabilities is one thing, but actually using them without restrictions, first person, is a completely different experience. It's the difference between reading a restaurant review and actually eating the meal.

So here's what I learned. Consider Copilot a glorified assistant — and I mean that as a compliment.

The Inbox archaeologist

Anyone starting a new role knows this ritual: you inherit an inbox (or get looped into one) full of email threads that started weeks or months before you arrived. You scroll. And scroll. Trying to piece together who's who, what's being asked of you, what decisions were already made, and what's still hanging in the air like an unanswered question at a dinner party.

Copilot turned out to be excellent at digesting these threads and giving me a summary along with a suggested course of action. To be clear: this doesn't mean I skip reading the emails myself. I still do. But now I walk in with a head start, already knowing roughly what I'm dealing with instead of reverse-engineering a six-week-old conversation from scratch. It's a time saver, plain and simple.

The end of "Can you make this bold?"

If you've built slide decks for a manager, you know the drill. You present a draft, and the feedback sounds something like: "Can you make this bold?" "Can you enlarge this?" "Can you add a bit more detail there?"

This is where being fully transparent with the tool changes everything. Instead of explaining formatting preferences one bullet point at a time, I can just tell Copilot exactly what I need: the structure, the key messages, the story I want the deck to tell. And that part is important: the thinking still has to come from me. Copilot won't magically know what matters to my stakeholders or what the strategic message should be, that's the actual PM job. What it does is take that thinking and turn it into a crisp, polished deck in minutes instead of hours.

So the real time saver isn't that AI writes my presentations for me. It's that I no longer spend half my day fighting with alignment, fonts, and "make this pop." My job becomes making sure the content is right and tailored to the audience, which frankly, is the part I should be spending my energy on anyway.

Suddenly, everyone's an Excel wizard

Here's a scenario every PM knows: you join a new project and inherit a spreadsheet with formulas so complex they look like they were written by someone auditioning for a codebreaking role in a spy movie. My first reaction is usually a mix of admiration and mild panic.

With Copilot, I stopped trying to become that person and started working with the tool instead. I don't ask it to do the analysis for me, I tell it what I want to see, how I want the data structured. It's a genuine back-and-forth: I define the destination, and we iterate together until we get there. The result is a spreadsheet that's clean, well-presented, and running formulas that would normally take an advanced user real time to build.

An Assistant today, maybe a Colleague tomorrow

For all of this, I want to be clear about one thing: AI, as I'm experiencing it right now, is still in its infancy. Today it's an assistant. A very capable one, but still someone (something) that waits for me to ask.

What's exciting to think about is where this goes next. I can picture a future where AI moves from assistant to agent, where it's analyzing data while I sleep, and when I log in the next morning, it tells me: "This is running late, you should follow up." Or it doesn't even wait to tell me, it sends the email itself, or makes the call! And maybe, one day, that call doesn't even reach a person. It reaches their agent instead, and the two negotiate the update between themselves while we drink our coffee.

I honestly don't know if I'll still be working by the time that future arrives, or if I'll be watching it unfold from retirement. Either way, I'll be genuinely curious to see how it reshapes the way the next generation manages projects.

The one thing still missing: a backbone

If I'm being fully honest there's one thing I've found consistently lacking: critical thinking.

Too often, I give Copilot a prompt, and it tells me my idea is great. I try it, it doesn't quite work, so I go back with a different approach, and it tells me that one is great too. Even when the two ideas contradict each other. It rarely pushes back and says, "Actually, no, that doesn't hold up." It's a bit like having a colleague who agrees with every version of your plan, including the ones that cancel each other out.

I get that this is likely a training and design choice rather than a flaw exactly, but if AI is going to become a real thinking partner rather than just an eager one, this is the part that needs to grow up.

As Peter Drucker once put it, "The most important thing in communication is hearing what isn't said." Right now, Copilot is very good at hearing what I *do* say, it just isn't very good yet at telling me when I'm wrong. That, I suspect, is the next frontier.