It’s a funny thing about us humans, particularly when we find ourselves in positions of leadership. There’s an almost gravitational pull towards control, towards centralising decisions, towards believing that if we just gather enough information at the top, we can somehow orchestrate everything perfectly. I’ve seen it countless times over my 40 years in the tech world, and if I’m brutally honest, I’ve felt the tug of that instinct myself in earlier days. It’s a comfortable illusion, control. But like many comfortable things, it often limits true potential.

It's a comfortable illusion, control. But like many comfortable things, it often limits true potential.

My physics background often leads me to look for underlying principles in complex systems, and businesses are certainly complex. One of the most robust principles I’ve observed, and one that became a cornerstone of how we operated at Constellation, is the almost paradoxical power of genuine autonomy. I say paradoxical because giving it often feels deeply uncomfortable, yet it’s one of the most effective ways to scale an enterprise, particularly one made up of diverse, niche software businesses.

This wasn't a revelation from a neatly packaged MBA module: I never did one of those. My education in this came from thousands of hours with business books, from observing countless companies, and, perhaps surprisingly to some, from experiences long before I ever sat in a boardroom. I spent my pre-university years doing a variety of labouring jobs: cleaning floors at Mr Kipling's factory, humping boxes in warehouses. These weren't glamorous, but they were profoundly educational. I saw, at the ground level, the immense difference between a team that felt ownership and one that was merely following orders from a distant office. I saw individuals with incredible ingenuity and dedication, often in roles that many would overlook. It instilled in me a deep respect for the capabilities of people when given a clear task and the space to get on with it.

Fast forward to the world of software acquisitions. When you’re building a conglomerate of many distinct vertical market software (VMS) businesses, the temptation for a central command structure is immense. "We need standardised reporting!" "We need a central R&D function!" "We should centralise sales and marketing for synergies!" The arguments are always couched in terms of efficiency and control. And on paper, they can look compelling. The problem is, software businesses, especially niche VMS ones, are often idiosyncratic. They serve unique customer bases with unique needs. A one-size-fits-all central approach often ends up being a one-size-fits-none, adding layers of bureaucracy and slowing down decision-making.

Autonomy in Practice

Consider a hypothetical software business we might acquire, let’s call it "MarinaMaster," a leader in software for managing boat marinas. The MarinaMaster team knows their customers, their competitors, the seasonal quirks of the industry, far better than anyone at a central head office ever could. If we, from a great height, started dictating their product roadmap or their marketing strategy, we’d almost certainly get it wrong. We’d demoralise a competent team and blunt their effectiveness.

The alternative, genuine autonomy, is to say to the MarinaMaster General Manager: "This is your business. Here are the broad financial and ethical guardrails. We expect a reasonable return on the capital invested. How you achieve that, within those guardrails, is largely up to you. We’re here to support, to share best practices from other businesses if you’re interested, to provide capital for good ideas, but you’re accountable for the results."

Now, this is where the "uncomfortable" part comes in for many managers, both at the central level and sometimes even for the newly empowered GM. For central management, you have to resist the urge to meddle, to second-guess. You have to trust that the person running MarinaMaster is capable. You have to accept that they might make mistakes, and that those mistakes are learning opportunities. You have to accept they might do things differently than you would, and that "different" isn't necessarily "wrong." Your role shifts from direct command to coaching, questioning, and providing resources. It is, in miniature, the same discipline that AI agent deployment is now surfacing in leaders who have never quite mastered it with people.

For the GM, suddenly the buck truly stops with them. There's no one upstairs to blame if things go awry. It can be daunting. But for the right kind of leader (intelligent, energetic, ethical), it's also incredibly liberating and motivating.

And why does this uncomfortable approach scale so effectively?
Firstly, it keeps the central overhead remarkably lean. You don't need vast departments at head office trying to manage hundreds of diverse businesses. Your central functions become more about capital allocation, governance, and fostering a network for sharing knowledge, rather than operational command.
Secondly, decision-making is faster and more relevant because it’s happening closer to the customer and the market. MarinaMaster can respond to a new competitor or a customer request without waiting for three layers of approval.
Thirdly, it attracts and retains great talent. Good people crave ownership and the ability to make an impact. A genuinely autonomous environment provides that in spades. They're not cogs in a large machine; they're skippers of their own vessels.
Finally, it builds resilience. If MarinaMaster has a tough year, it doesn’t cripple the entire organization, because dozens, or hundreds, of other BUs are operating independently.

During my later years at Constellation, much of my time was spent coaching Business Unit leaders, helping them think through their strategies, including preparing for disruptions like AI and, increasingly, the adoption of AI agents that themselves require exactly this disposition toward delegation. This coaching model only works effectively if those leaders have genuine autonomy. You can’t effectively coach someone on navigating a complex market if they don’t have the authority to actually steer the ship.

The Same Lesson With AI Agents

I have noticed, in more recent conversations, that leaders who have never fully developed the psychological muscle to let go of control over people face an almost identical challenge when it comes to deploying AI agents: systems capable of making decisions, drafting communications, and executing multi-step processes with a degree of autonomy that requires genuine trust to be meaningful. The temptation is to insert approval checkpoints everywhere, to monitor every output so closely that the efficiency gains are effectively reversed. The dynamic is entirely familiar: it is the same instinct that turns a potentially effective leader into a bottleneck in a human organisation. In both cases, the answer is the same: clear guardrails, accountability for outcomes rather than methods, and the willingness to allow the system, human or otherwise, to do its work. The psychological shift required is not a new one. What is new is the speed at which the cost of failing to make it now compounds.

The journey to embracing true autonomy isn't always smooth. It requires a culture of trust, clear accountability for results (not methods), and a willingness from senior leadership to genuinely let go. It means accepting a degree of "messiness" that can make traditional, hierarchical managers uneasy. But the payoff, in terms of sustainable growth, innovation, and human potential unleashed, is immense. It’s about fundamentally trusting good people to do good work, a lesson I started learning long ago, not in a lecture hall, but on the factory floors and in the warehouses, and one that has proven its worth time and time again in the complex world of software. It might be uncomfortable, but its power to scale is undeniable.