You got out of the dugout. You started with the work.
Maybe you made a spreadsheet. Maybe you talked with your team. Maybe you documented three AI uses you knew about and six you didn’t.
Either way, you looked. And now you know where AI is showing up.
This is where organizations get stuck.
Because seeing the work is not the same as deciding what to do about it. A list of AI uses isn’t a decision. It’s an inventory. And an inventory, left alone long enough, can start to feel like permission, not because anyone granted it, but because nobody said otherwise.
The next question is:
Now that we know what’s happening, what are we willing to stand behind?
An inventory documents what’s happening. A practice guideline defines what you’re accountable for.
This is where experimentation starts becoming practice.
Pick one thing. Not everything.
Once organizations see the full inventory, there is a temptation to govern all of it at once.
Write the policy. Cover every possible use. Get the language airtight. Send it to the board. Done.
Except AI doesn’t seem particularly interested in staying done.
You don’t have to figure out every future AI use before you can govern the one happening today.
Pick one.
Something real enough to matter, but contained enough to learn from. Meeting transcription. Grant summaries. Communications drafts. Media monitoring.
One foundation I worked with picked a workflow staff were already experimenting with. Rather than starting with a universal rule about AI-generated analysis, we could ask what responsible use meant for that particular practice.
What sources should the system rely on? How would results be checked? Where did staff judgment matter? Who owned the work? And, importantly, what would make the output useful enough to keep doing?
That’s a much more manageable governance problem than:
What is our position on AI?
And it produces something more useful too.
It produces a decision.
From “we’re trying AI” to “this is how we do this”
A practice guideline is simply the written record of that decision.
One page, maybe two.
For a particular piece of work, I want an organization to be able to answer five things.
Purpose: What are we trying to make better?
Not “use AI.” What should actually improve?
Time saved may matter. So might better analysis, greater consistency, increased staff capacity, stronger relationships, or a better experience for the people on the other side of the work.
I’ve been thinking about this even more after listening to ServiceNow Chief People and AI Enablement Officer Jacqui Canney talk about AI adoption and impact. The conversation kept pushing beyond whether people are using AI, or even whether it makes them faster, toward a better question:
What changed because of it?
That distinction matters.
I’ve argued elsewhere that efficiency is the floor. The more interesting question is what that efficiency makes possible.
A practice guideline shouldn’t only tell us how to use AI safely. It should remind us why we’re using it at all.
Ownership: Who is responsible for this practice?
Not merely who has the license or writes the prompt. Who is responsible for whether the practice works as intended?
A name is more useful than a department.
Ownership doesn’t mean one person has to understand every technical, legal, operational, or ethical dimension of the technology. It means someone is responsible for making sure the questions get asked and the right people get involved.
Boundaries: What can AI touch and do?
What information is involved? What can the system produce? What could go wrong? What isn’t it allowed to do?
Public information behaves differently than internal organizational data. Grantee information raises different questions again.
Say the actual failure out loud.
Maybe a summary misrepresents a grantee. Maybe confidential information gets entered into the wrong tool. Maybe a polished draft goes out under someone’s name without meaningful review.
Govern the risk that actually exists in the work, not every imaginable AI catastrophe.
Human judgment: Where does a person remain load-bearing?
This is one of my favorite questions.
Where in this practice do context, expertise, relationship knowledge, interpretation, or judgment actually matter?
“Human in the loop” sounds reassuring until nobody can explain what the human is supposed to be doing there.
Evidence: How will we know whether this is working, and for whom?
Not just: Did anyone complain?
And not just: Did it save time?
Did the quality improve? Did staff gain meaningful capacity? Did people make better decisions? Did the experience improve for the people the process is meant to serve?
Sometimes the evidence tells you to expand the practice. Sometimes it tells you to change it.
And sometimes the answer is:
We’re not ready to do this yet.
That’s a governance decision too.
A practice guideline doesn’t exist to bless every AI experiment. It helps an organization decide which experiments should become practice, under what conditions, and which shouldn’t.
The practice matters more than the tool
Another funder I’ve been learning alongside encountered this from a completely different direction: AI-generated grant applications.
It would be easy to center the question many funders initially asked:
Was this written by AI?
That question is becoming less useful by the day.
Instead, the organization focused its proposal review guidance on something more meaningful:
Is this a good application?
Applicants could use AI to support their writing. But they remained responsible for providing specific, accurate, responsive answers grounded in their organization’s actual experience.
That’s a subtle but important governance choice.
It shifts the focus from policing the presence of AI to governing the quality and accountability of the work.
I’ve watched similar organizations draw a related line internally. AI can help staff organize information, identify themes, and work through large volumes of qualitative material faster than a small team could manage alone.
The boundary I’ve seen work best is simple:
AI organizes and surfaces. It doesn’t make the funding decision.
People remain responsible for engaging with applicants’ stories, understanding context, and exercising judgment.
Use the tool. Don’t lose the relationship.
That’s a practice decision. And once an organization can say it that clearly, something changes.
The decision becomes something you can document, teach, revisit, and apply to the next workflow, instead of a feeling everyone shares but nobody has written down.
Write down what you’ve learned
This is the part I think organizations underestimate.
Not because they resist governance. Most of the leaders I work with take it seriously.
But writing it down can feel like an extra step after the real work of deciding is already done.
It isn’t.
Experimentation produces organizational knowledge. Without a way to capture it, that knowledge evaporates. The same lessons get relearned by the next team, on the next tool, six months later, from scratch.
A good practice guideline holds that knowledge in place.
It says what the practice is for, who owns it, what information is appropriate to use, which parts AI can support, and which parts stay squarely with a person. It names the boundaries, what happens when something falls outside them, and how the organization will know whether the practice is producing what it intended.
It turns “we’re experimenting” into “we made a choice.”
It gives staff something sturdier than “use good judgment.” It gives leaders something more honest to say than “I think we’re being careful.”
And it gives the organization something to hold up against the technology when it changes again in six months.
Because it will.
The organizations making real progress aren’t treating that change as a governance failure. They’re updating the guidance, testing the next practice, and adjusting once more.
That’s governance working.
This is how governance starts to become real
One practice guideline doesn’t make an organization AI-governed.
But do this with three or four meaningful uses and patterns start to emerge.
The same data concerns surface across different practices. The same human decisions keep turning out to matter. You start to see where staff need more training, which questions need technical expertise in the room, and which risks can be handled within a practice versus which require an organizational decision.
This is the muscle: the reflex to write down what you’ve actually decided when a new use, tool, or question shows up, rather than noticing that things seem to be going fine and moving on.
That’s what second base looks like from the inside.
Not a governance framework arriving fully formed and handed down from above.
A growing stack of specific, written guidelines. Each one owned by a real person. Each one describing real work your organization is actually doing.
You don’t round second by declaring your values as they relate to AI-usage.
You round it by writing down, one practice at a time, what you’re actually willing to do, and asking the question that keeps all of it honest:
Is it still working?
You already found the work. Now pick one thing, and write down what you’ve decided.
Ready to write your first practice guideline?
💬 You’ve done the inventory. Maybe you’ve done the training or policy work too. The next step is picking a real use and making the call.
I work with COOs, executive leaders, boards, and grants management teams to turn AI experimentation into documented practice: clear purpose, ownership, boundaries, human judgment, escalation, and evidence for real work already happening inside the organization.
If that’s the gap you’re sitting with, I take on a limited number of new engagements each quarter. Q4 is filling.
➡️ Let’s talk.
About the Author
Rachel Kimber is managing director of Full Circle Impact Solutions, where she helps foundations and nonprofits build the infrastructure to use AI effectively, ethically, and in service of mission. She works with boards, leadership teams, and grants management staff to align technology adoption with institutional values.
She is faculty for PEAK Grantmaking’s Practical AI for Grants Management Learning Series and NTEN and TAG’s certification programs. Rachel co-created the Stuck to Strategic Framework and speaks internationally on AI, governance, and organizational strategy.