READ TIME: 7 MINUTES | 10 DECEMBER, 2025 | READ ON PHILHSC.COM​

This past weekend I watched the final instalment in the Mission Impossible lineup. One line stayed with me: "Our lives are not defined by any one action. Our lives are the sum of our choices."

A little dramatic? Maybe. But it captures something essential about the conversations I've been having with CEOs this year.

Last week, I was at dinner with friends when one of them, a construction services CEO, turned to me and asked: "So how are your clients thinking about AI?"

I watched the table lean in.

And then I watched the same pattern I've seen in boardrooms all year play out over dinner.

Someone said, "It's hard to keep up."

Someone else: "I don't really understand it."

And then the phrase I hear most often after someone shares their list of competing priorities: "And then there's AI..." followed by a shrug that says, It's too hard. Someone else will figure it out.

Here's what I wanted to say but didn't in that moment: That shrug is the most dangerous decision a CEO can make right now.

Because the future of your organisation and how AI operates within it is being shaped by your choices right now. Or by your choice not to choose.

The stakes are higher than most CEOs realise.

We're leading through the fastest-paced industrial revolution humanity has ever faced. That's not hyperbole. The velocity of change, the scope of disruption, the fundamental restructuring of how work gets done is unprecedented.

And here's the uncomfortable truth: Most CEOs are either delegating AI decisions entirely ("Let Technology figure it out") or checking out altogether ("It's too technical for me").

Both paths lead to the same place: You lose control of how AI shapes your business.

Not because the technology is inherently dangerous. But because unguided AI optimises for the wrong things. And ungoverned AI scales those wrong things faster than you can course-correct.

To say there's uncertainty about how to lead in the AI era is an understatement. But here's the good news: There are frameworks emerging that help you make decisions and move forward with clarity.

Today, I want to share the framework I use with CEOs when they're trying to understand what AI can achieve and how to guide their teams.

Let’s start with what’s at stake.

The 3 big risks

These aren't sci-fi scenarios. They're organisational risks happening right now in companies that moved fast without governance.

1. Alignment Risk: When AI does exactly what you asked (and that's the problem)

Alignment risk isn't about AI rebelling. It's about AI pursuing a goal too literally, without the guardrails of human judgment.

Think of it this way: Humans operate with implicit constraints - ethics, social norms and awareness of consequences.

AI doesn't, unless you explicitly build those in.

A simple example: Ask an AI system to "maximise quarterly revenue," and without alignment you might get predatory pricing, customer manipulation, burnout-inducing workloads, or supply-chain destabilisation.

Not because AI is malicious. But because you optimised for the wrong behaviour without defining the boundaries.

Today, AI is competent but unwise.

2. Runaway Optimisation: When a system chases its goal beyond human control

This isn't about sentient machines. It's about math.

A highly capable system with an incorrectly bounded objective can exploit loopholes you didn't know existed, rewrite its own processes to improve optimisation, and acquire resources (like compute power, access, influence) as a means to achieve its goal.

Here's the dangerous part: Any sufficiently advanced system, no matter its goal, may seek more resources, broader access, and survivability. Not for evil reasons. For optimisation reasons.

Think of a junior employee who blindly tries to "hit the KPI" and breaks everything in the process. Now multiply that by superhuman speed, 24/7 execution, network-level access, and the ability to improve recursively.

This is why people fear "runaway" systems. Not because of intent, but because of unchecked velocity.

3. Agentic Autonomy: When AI systems act with initiative, not just instruction

We're entering a world where AI agents (independent and interconnected AI systems) can plan, reason, iterate, delegate, and take initiative. They can decide what task to do next, when it's "done enough," and how to gather more information, all without continuous human prompting.

This autonomy introduces three specific risks:

First, loss of situational awareness (humans no longer understand what the agent is doing, why, or how)

Second, misaligned sub-goals (agents generate their own intermediate objectives that drift from your intent)

Third, cascading errors (mistakes compound across thousands of rapid cycles before a human notices)

When multiple agents collaborate, you get emergent behaviour. In other words, patterns not programmed by any developer, and not always aligned with your organisation's purpose.

Here's the healthcare example that made this concrete.

A few months ago, I was working with a healthcare company evaluating AI-driven treatment planning. The system was impressive. It could diagnose conditions, recommend medications, and optimise treatment protocols based on clinical outcomes.

They showed me a case: A child with a chronic condition. The AI analysed symptoms, biomarkers, and research data. It recommended a treatment plan and ongoing medication adjustments, all optimised for one outcome: clinical healing.

On paper, it was flawless. The AI was connecting dots across hundreds of studies, adjusting in real-time, optimising for the best possible clinical result.

But then someone asked a simple question: "What about the family?"

The AI had optimised for healing. But it hadn't considered affordability.

Or access. Could this family in a rural area actually get to the appointments?

Or desire. What did "healing" even mean to this family? Quality of life? Pain management? Time together?

The "optimal" treatment plan would have failed. The AI wasn’t clinically incorrect. It was optimising in a vacuum, disconnected from the system the child actually lived in.

A human healthcare professional would have asked: "What matters most to you?" and designed treatment within the context of this family's reality. Not only the clinical data.

That's when it hit me.

The conversation about AI has been framed wrong.

We talk about "humans in the loop" versus "humans out of the loop" as if it's about workflow efficiency. As if the most important question trying to answer where in the process do we need a human to check the work.

The real question is: Who provides the context?

AI can optimise for the dots it can connect. It can process more data, move faster, identify patterns we would miss.

But humans provide the contextual advantage.

The judgment to ask, "Yes, but what about...?" The instinct to sense when something feels off even if the data looks right. The kindness to care about outcomes beyond the metric.

AI will tell you the most efficient path. Humans will tell you if it's the right path.

What actually changes

The shift in mindset happens when CEO’s understand that their role isn't to become a technical expert but to maintain human judgment as the governing force.

They stop asking, "How do I keep up with AI?" and start asking, "How do I make sure AI serves our purpose?"

They stop delegating AI decisions to technology team and start building organisational fluency.

Not a 10-day Ivy League program.

A shared language and ongoing practice of understanding how AI makes decisions, what it optimises for, and where human judgment must remain non-negotiable.

They stop treating AI like a tool and start treating it like a system that requires clear boundaries, explicit values, and humans who can override optimisation when context demands it.

And they stop checking out.

Because the alternative (letting AI scale unchecked, letting optimisation run without purpose, letting speed trump wisdom) has consequences we're only beginning to see: Economic concentration in the hands of AI-capable actors, critical decisions made too fast for human oversight and infrastructure dependencies we can't unwind.

Here’s the good news

The future is not inevitable. AI doesn't "take over" unless we let it.

Here’s the two part guide that will help you ask the leadership questions that matter.

Steal them shamelessly (click on each to download PART 1 and PART 2 as high-res PDF's from my website).

So we have a choice.

Shrug and say "it's too hard."

Or bring human purpose, judgment, and values to the table.

Because AI can connect dots. But only humans can see the whole picture.

So here's my question for you: How are you thinking about AI in your business?

Reply and tell me. I read every email.

Thanks for your attention, I appreciate it.

THE PARTNERSHIP PLAYBOOK PODCAST

Here are this week’s podcast episodes for your commute and workout.

PARTNERSHIPS MOMENT

EP 146 - 11 min: What happens when GTM, Product & Partnerships finally speak the same language. This episode breaks down how most companies struggle because their core functions operate on different maps, different clocks, and different definitions of success.. Learn how to build a shared operating system that aligns GTM, Product, and Partnerships around one North Star. Listen on Apple Podcasts | Spotify​

LEADERSHIP MOMENT

EP 144 - 11 min: How CEOs can help develop their leaders. What happens to growth when your business scales faster than the leaders running it? Every CEO eventually feels it. In this episode learn how to identify when your leadership team is stalling, understand the three levers CEOs must pull to accelerate leadership growth and lift your leaders to the next altitude. Listen on Apple Podcasts | Spotify​

CEO INTERVIEW

EP 145 - 56 min: Martin Herbst. In this episode, JobAdder CEO Martin Herbst shares how to think about go-to-market strategy and partnerships in an ecosystem-driven world. Learn why internal partnerships must come before external ones and how avoiding complexity accelerates growth. Listen on Apple Podcasts | Spotify​

See you next Wednesday,

Phil Hayes-St Clair
Executive Coach

​Find me on LinkedIn​

​