This year, everyone in our sector has been getting the same question: "What's your AI strategy?"
Here’s mine: get the people who already know our work to use AI tools, with real access and encouragement to thoughtfully experiment. That’s the headline. It’s simple and deliberate, and I’m confident it’s actually the most mature answer. It pays off quickly, and it keeps compounding. The organizations and teams that focus on this habit now will be well ahead of the ones still waiting for a plan.
The most valuable thing AI can do for most orgs in our sector right now isn't to replace anyone, and it isn't to run on its own (two topics I can't wait to dig into in the coming weeks). It's nowhere near that yet, and even if it were, most of us aren't set up to use it that way. Its real value in 2026 is simpler: it lets someone who already knows the work do more, and sometimes do it better. But that only shows up when those people know when to use it and get good at using it.
Here's what makes right now the right time for this approach. The tools most of you already have - Copilot inside Microsoft 365, or Gemini in Google Workspace - can now see your actual work: your documents, your email, your messages to colleagues, your data. That context makes them genuinely useful. It does not make them reliable.
You'll get a moment where it pulls a messy set of email and message threads into a clean summary, in your own voice, folding in exactly the context it needed. It's a little startling. Ten minutes later you'll ask it for the key figures behind a budget, and it hands you something confidently, completely wrong. Some of that is the data pipeline underneath (a topic for another day), but it's also a sign of how "jagged" these tools still are - sharp on one task, badly off on the very next.
That whiplash is an important lesson. The tools are finally decent enough to be worth your people's time, and still flawed enough that using them helps build real judgment and understanding that will be critical moving forward as an org-wide strength as the tools advance.
It’s not just this trust lesson that’s important to seed and experience. It’s that human expertise is the missing ingredient that AI may never solve for. Tools can generate all day, but they can’t know what “good” looks like in your world, for your stakeholders, or for where the mission goes next. That judgment belongs to you and the people you trust with critical work, and it cannot be handed to a model. Encouraging repetitions will sharpen how your teams can validate outputs, but also give them real experience in understanding how outputs are arrived at.
Let’s take where this conversation was last year. The typical focus for getting staff on-boarded into AI was all in on “prompt engineering” - teaching staff a standard recipe to write prompts that encouraged high quality output.
Now we have tools and models that are multi-modal, rely less on the precision of the prompt, and attempt to infer from the context can access to determine what is required. They have fledgling memories, engage in conversational back-and-forth to chain together reasoning loops, and can be personalized so that output is shaped to the way the user may prefer or require.
This is a sign of what's coming, and it's why exposure right now is so important. Outputs sound increasingly confident and relevant even with weaker (or lazier) prompting, but they can still be poor. The confident tone is what makes the weak ones easy to miss, and teams need practise to see where things are thin vs genuinely impressive.
The input recipe becoming richer is also critical to think about. It used to be about the prompt; now it's about knowing what range of inputs and context produce a good output, or where strength might come from scaffolding together individual pieces in turn, and that's different for fundraising than it is for finance. It can't be trained effectively just from the centre; it has to be learned in the seat, and the people in that seat are the ones who'll start to spot where AI is genuinely worth using in their work, and where it isn't. Giving your people at-bats doesn't just make AI use safer and more effective in the short term. It builds a team that can identify where the opportunity is to move the mission or change the workflows as the tools get stronger and better suited to the work.
This isn't a new file share, a new fundraising platform, or a new set of brand guidelines, where a couple of training sessions get everyone up to speed. The trajectory of these tools points at something larger: a real change in how the work itself gets done, in nearly every role you have. So exposure can't be a pilot, and it can't be one more thing that waits on a strategy and a scoping doc. Treating a shift this size as a tidy project to be managed is the actual risk.
So, The Thing Is. Your job isn't to write the master plan or chart the course, and it isn't to tell people to go play with AI. It's to set up the conditions so they can actually get going, start a valuable hands-on education, without creating unnecessary risk. A few things can make that real:
Give people genuine access. Turn on one tool properly (most of you already have Microsoft Copilot in your Office licensing, or Gemini in Google Workspace), so nobody's slipping sensitive work into a personal account to get around a “wait for our plan” hold. These are good choices with strong models, and likely already protect your data under your existing agreements.
Pair it with a short dos-and-don'ts: humans stay in the loop, no private or donor data, no blind trust in the output, and don't assume anything you type stays private. (I keep a one-pager for exactly this. Reply and I'll send it.)
Then do the part that actually takes leadership: build a culture where experimenting is encouraged and saying "that didn't work" is just as welcome, because the learning lives in both. High performers will be wary that they might be judged for using AI. Make it clear that it’s a positive to see AI in their work, not a negative. And hold one line without exception: nothing an AI touches goes out the door until someone who knows the work has checked it. That validation habit is what turns loose experimentation into real judgment instead of risk.
The Bottom Line
At-bats are the strategy, and they are the best one in 2026. Putting AI in the hands of people who already know the work, and letting them build judgment and skill by using it, is the fastest and most defensible path to value, this year and well beyond.
Your job is the conditions, not the plan. Real access to one safe tool, a short dos-and-don’ts, a culture where experimenting and failing are both safe, and one hard rule: a person who knows the work validates every output.
Start this week. Ask your IT team what it would take to turn the tool on, share the ground rules, and have a few people run something real, then compare what nailed it against what fell flat.
The next time someone asks what you're doing about AI, that's your answer.
Thanks for reading the first issue of The Thing Is. I started this because our sector is drowning in AI hype and starved of anything that actually speaks to how we work, what we can afford, and what we're accountable for. I want this to be a straight, useful voice in that gap; and a two-way one. If something here landed, or missed, or there's a question you're sitting with, I genuinely want to hear it. This gets better the more I hear from the people I'm writing it for. Thanks again.
— Mike