September 3, 2026

A short time ago, I wrote about something that had been brewing inside my brain. As AI works its way into more of how we analyze information, how are we making decisions? In particular, impactful decisions at an organizational level?

AI tools continue to improve at collecting information, finding patterns, organizing what we know, and helping us evaluate options. As mentioned in the last Playbook Notes, I use them every day.

What they don’t do is remove the need for judgment. Someone on your leadership team still has to decide what’s true, what matters, which tradeoff is acceptable, and what happens next.

Since writing that piece, I’ve had a number of conversations that pushed the question even further.

The more interesting issue may no longer be how individuals use AI. It may be how firms are beginning to operate around it.

From tool to operating environment

Most of the early conversations about generative AI were about tools. Which model are you using? Can employees use ChatGPT? Should we buy Copilot? Can it write this faster? Can it summarize that?

Those were reasonable questions. In many ways, they still are. But AI is increasingly moving into the places where work already happens. Places like meetings, your CRM, investment research, firm and team member communications, and compliance review systems. But it is getting even deeper now. Client service tools. Knowledge systems. Task management and workflows. Even automated client experience platforms.

Sometimes, your team will deliberately choose to use AI. Other times, AI simply arrives without much fanfare as a new capability inside software the firm already uses. And that changes the nature of the conversation.

Once AI is embedded in the actual flow of work, the question isn’t simply, “Which tools should the firm use?” It becomes, “What role will AI have inside the firm?”

Fundamentally, the big question is whether leadership actually made that decision intentionally and strategically.

And that is where empires will rise and fall. Or, at least, firms may excel and grow versus playing constant catch-up in a compounding failure mode.

You may already have an AI Operating Model

I suspect most organizations will eventually have something that could reasonably be called an AI Operating Model. Many already do. They just didn’t set out to design it as one.

The story usually goes something like this. One department buys a tool. Another activates an AI feature inside an existing platform. Someone starts recording meetings. Team members develop their own prompting habits. A vendor adds summarization. Compliance and legal establish a policy. IT sets permissions. Operations adapts the workflow.

Every one of those decisions may make perfectly good sense on its own. Taken together, though, they begin to define how AI operates inside the business.

What can it see and retain? What can it interpret or create? What can it initiate? Where does human review belong? Who owns the result when AI touches the process?

Those are no longer just technology questions. They are operating questions.

Today, firms are already answering them whether leadership realizes it or not.

The questions get harder as the tech gets better

Consider something as ordinary as a meeting. AI can already record the conversation, transcribe it, summarize what happened, identify action items, capture decisions, and organize follow-up.

That’s useful. I use those capabilities myself. But it doesn’t take much imagination to see where this is headed.

A system that carries context from one meeting to the next can begin to remember prior commitments, surface unresolved issues, recommend next actions, and preserve the history behind a decision. At that point, we’re not really talking about better meeting notes anymore. The meeting starts becoming one of the places where the firm’s operating system gets updated.

That raises more consequential questions. What should the system remember? Who decides whether its summary reflects what was actually decided? At what point does recommending an action become initiating one? And what happens when the system is confidently wrong?

I think some of what is being built right now could materially improve how organizations make decisions, preserve context, communicate, and follow through. That is precisely why these questions matter.

“Human in the loop” doesn’t answer very much

One default response to all of this is that a human will “remain in the loop.” Great. But which human? At what point? And what is that human accountable for (and to whom)?

If AI summarizes a leadership meeting and that summary becomes the record everyone relies on, someone still has to determine whether it captured the decision rather than merely memorializing the conversation. If AI recommends the next step based on months of prior context, someone still has to recognize when an important, or more recent, piece of context is missing.

Generic “responsibility” can sound reassuring without actually defining ownership or decision criteria.

The answer cannot simply be “with the human.” The failure rate on that default is predictably high. Everyone’s job is no one’s job. Organizations still need to decide where human judgment belongs and who owns the consequences of each decision.

Your vendors are choosing some of this for you

Another aspect of this is easily overlooked. Software has settings, permissions, defaults, retention rules, integrations, and automated actions. Increasingly, it also has embedded intelligence and machine learning tools.

Those realities are often treated as something inside technical configuration. Some of them are. But some are not. A setting that determines what gets recorded, retained, shared, or acted upon may also determine how the organization operates. Without thinking too deeply about it, you may “default” into a habit no one thought about.

At what point does configuration become an operating decision? The lack of an answer today doesn’t mean a vendor is doing anything inherently wrong. What it does mean is that leadership needs to recognize, and document, the difference.

Otherwise, a firm’s AI operating model can become the accumulated result of software settings, departmental purchasing decisions, employee behavior, and whatever seemed convenient during implementation, with compliance and legal stepping in at key inflection points to manage risk and reputation.

That’s still an operating model. It just isn’t necessarily the one anyone consciously or deliberately chose.

Policy matters. It isn’t the same thing.

Most firms taking AI seriously are also developing policies around approved tools, confidential information, security, supervision, and acceptable use. They should. But policy and operating design are not the same thing.

A policy establishes boundaries. An operating model defines how the organization intends to work when AI is part of the process. You can have a thoughtful AI policy and still have several different answers to that question across the same firm.

The same distinction applies to AI strategy. AI strategy may define where the firm wants to invest, which capabilities matter, and where the technology can create value. An AI Operating Model addresses what happens once those choices become part of everyday work and firm structure.

Both are important, but they answer different questions.

Institutional memory changes the stakes

An interesting possibility emerging from this technology is the ability to preserve institutional memory in a way that has historically been very difficult.

Organizations lose context constantly. Important pieces of a decision wind up spread across meetings, emails, CRM notes, documents, Teams or Slack conversations, and individual memory. People leave. Roles change. Months pass. Eventually someone asks why a decision was made, and the answer depends on who still remembers the meeting. AI has the potential to materially improve that.

A system that can connect discussions over time, preserve decisions, surface prior commitments, and bring relevant context into a current conversation could solve a very real operating problem. It could also make meetings much more useful because the value of the conversation would not disappear when the call ends.

But institutional memory isn’t simply a storage problem. Someone still has to decide what deserves to persist, what represents an actual decision, and what remains unresolved. There is also a meaningful difference between what was said, what AI interpreted, and what leadership actually decided.

What happens when today’s imperfect record becomes tomorrow’s source of truth? Once AI begins using institutional memory as an input into future analysis and action, errors will not necessarily stay in the past. They can travel forward undetected and become institutional without intent.

The promise is enormous. So, too, is the responsibility to design it well.

A thoughtful AI Operating Model can make good policy and practice durable while reducing the risk that problematic defaults sneak into your ecosystem.

Before the model becomes invisible

None of this means firms should stop experimenting until every possible question has been answered. The technology is moving quickly, and useful capabilities are becoming available now. Firms should learn where they genuinely improve the work. But speed and intentionality are not opposites.

There is a point where an experiment becomes normal operating behavior. A feature becomes part of a workflow. A summary becomes part of the record. A recommendation starts influencing action. That is when leadership needs to recognize that something more consequential has happened.

You do not need to answer every future AI question today. You probably couldn’t if you tried. But some questions are already important enough that leaving the answers scattered across software settings, policy documents, vendor agreements, individual habits, and meeting conversations creates its own risk.

The technology will continue to change. The firm’s operating decisions still need to be deliberate.

The harder question is whether leadership will define those decisions before the model becomes so embedded in the business that no one remembers making them.

And when those decisions matter that much, they probably shouldn’t live only in people’s heads.

William J. Cerynik | Founder and Principal, ApexEdge