The people I know who are genuinely good at working with AI tend to be project managers or product managers. I see it across both large organizations and mission-driven nonprofits. They were already strong at documenting things. They already knew how to make the right kind of request, whether that request was going to a person or an AI agent. The tool arrived and they were, more or less, ready for it.
That pattern says something the AI conversation mostly skips over.
Why AI maturity isn’t a tech skill
The conversation around AI tends to focus on the technical layer — model selection, tooling, and for a while, prompt engineering. A few years ago, prompt engineering genuinely mattered. You had to tell a model to role-play as an expert copywriter before it would give you anything usable. That era is essentially over. The current crop of models is conversational enough that you can just ask the question.
What’s replaced prompt engineering looks a lot more like good management — closer to onboarding a new hire than configuring software. A new collaborator needs context, clear expectations, institutional memory, and a specific picture of what a successful result looks like for your organization. So does the tool. The organizations that already knew how to provide those things are the ones pulling ahead now.
The underlying skill is older than any of this technology. It’s knowing what you want before you ask for it.
What this looks like in practice
In our own practice, we’re spending more time coaching client teams, helping them get better at communicating the outcomes they want rather than just running standard processes.
Take an agile development workflow. On an AI-enabled team, the ceremonies that carry the most weight sit at the two ends of the cycle: the elaboration of business requirements — getting genuinely clear about what you need the output of a project to be — and the retrospective, where you figure out what worked, what didn’t, and what you’ll do differently next time. The daily check-ins matter less than either one. Clear expectations going in, honest evaluation coming out, and a commitment to iterating in between, because this is never a one-and-done situation.
The same discipline applies to organizational memory. You get consistent results when you build a memory system that gives your AI agent the context it needs — everything you and it have learned through multiple cycles of working together — so it comes into every new interaction understanding what you expect and what a correct output looks like for you specifically. A lot of organizations haven’t done this, because their knowledge lives in people’s heads, in email threads, in whoever has been there the longest. AI makes that fragility visible fast. It also makes closing the gap worth the effort in a way it never quite was before.
The restaurant flyer problem
With AI, you have to be precise about what you want, and vocal about not accepting the first pass of whatever you get back. That first pass will almost always be generic. It’s one of the reasons there’s so much backlash against businesses using AI in their marketing. When teams accept the first thing the tool sends back, the result looks and feels like every other piece of output their competitors are also producing. It’s why so many restaurant flyers look identical on social media right now. Same layout, same language, same forgettable energy.
The homogenization is what happens when someone asks a vague question, gets a vague answer, and ships it. The tool did exactly what it was asked to do. Nobody in the loop had a clear enough picture of success to reject the first draft, because there was no standard to measure it against.
Your organization’s distinctiveness survives AI adoption only when someone holds a clear enough vision of the outcome to push back on the generic version. That’s a management function, and it always was.
Examine the foundation before you build on it
For mission-driven organizations, the stakes here are real. The communities you serve depend on your communication being specific and true to who you are. Generic output erodes that trust gradually, one interaction at a time.
Before layering AI into your operation, look honestly at the ground you’d be building on. How well does your team document decisions today? Could a new hire understand your expectations from what’s written down, or would they need months of hallway conversations to absorb it? Do your projects start with a specific, shared definition of success, or do people figure out what they wanted after they see it in the design phase?
If those answers make you uncomfortable, that discomfort is worth something. An AI tool placed on top of unclear communication produces unclear output faster and at greater volume.
Every one of these capabilities is buildable. And building them pays off whether or not you ever adopt another tool. Clear requirements, honest retrospectives, and captured institutional knowledge make your human team better too. The organizations seeing the most benefit from AI were already practicing some version of this. There’s no reason you can’t start now.
So this week, take one process your organization runs regularly and write a one-page description of what a successful outcome looks like — specific enough that someone unfamiliar with the work could evaluate the result. Have your team read it and tell you what’s missing. Do that before you evaluate a single AI tool.
