AI Won’t Make Leadership Less Important. It Will Make Bad Leadership More Expensive.

Leadership in the age of AI. Does it change, or does it just enable good leaders and find out the bad ones?

We’ve spent an enormous amount of time discussing how AI is going to change the work people do, not least in software development.

I’m increasingly interested in a slightly different question. What happens to leadership when AI changes the nature of the team itself?

My thinking on AI in software development has shifted considerably over the last couple of years, and most significantly in the last six months.

Initially, like most people, I saw AI primarily as a coding assistant. It could help a developer write a function, explain unfamiliar code, generate some tests or save half an hour searching Stack Overflow. Useful, certainly, but essentially just another productivity tool in the developer’s toolbox.

I don’t think that description is remotely sufficient any more. We’re moving from coding assistance (adding capacity to a developer), through individual augmentation (adding capability to a developer), and increasingly towards team augmentation (adding capablity and capacity to a team).

That distinction between assistance and augmentation matters

Those of you who watched my podcast with Mark Strefford will have heard about the work he has been doing in this team augmentation space, and it's a real window into the future of software development.

That distinction between assistance and augmentation matters, because if AI simply makes individual developers faster, it's a productivity question. If AI changes what an entire software team can do, it's an operating model question.

And if the operating model changes, leadership has to change with it, doesn't it?

What does this mean for leadership clarity?

More capability means more responsibility, and more things to be accountable for and worry about.

Imagine a software team that could comfortably deliver ten meaningful units of work in a given period in the old pre-AI ways of working. The leader of that team gives them a slightly poor steer. They misunderstand the customer problem, prioritise the wrong thing or make a questionable architectural decision.

That isn’t good, obviously, but the damage they can do is constrained to some extent by the capacity of the team. There's only ten things that can be wrong.

The general ability to create more "stuff" amplifies good decisions and bad decisions alike

In a team with AI augmentation, that same team can produce, say, five or ten times as much "stuff". What happens when the direction is wrong? You don't just get more productivity. You get the ability to be wrong much faster and at greater scale.

That, for me, is one of the most important implications of AI for leadership.

The general ability to create more "stuff" amplifies good decisions and bad decisions alike. The better AI becomes at executing work, the more important it becomes that somebody is setting the right direction in the first place. Being clear on the problem to be solved and the outcomes we're trying to achieve, what our constraints are and the things we deliberately choose not to do.

Those are leadership questions, not coding questions. And AI augmentation probably makes them more important.

Accountability doesn't disappear

Accountability is a topic I touch on with Mark in that podcast. It doesn't disappear when execution is delegated. I'd argue that accountability still has to sit with a human being. Ultimately, someone has to own the outcome.

There is an attractive idea buried within some of the discussion about autonomous AI agents. If AI agents can increasingly do the work, perhaps humans gradually have less to do? I don't think so.

There still needs to be a human making sure the right problem is solved

Delegation of execution has never meant delegation of accountability. At least, I don't think so.

In the old world, a leader doesn't stop being accountable for an outcome because somebody in their team performed the work. So why should that be any different because an AI agent performed some of it?

There still needs to be a human making sure the right problem is solved, in an architecturally robust way and with the right outcomes for the customer or end consumer. In fact, as execution becomes easier, I'd probably argue that accountability becomes more important rather than less.

How do we check progress?

There is another leadership habit that AI is going to challenge. The ways in which we check progress.

Traditional metrics around how many tickets have moved, how many points or stories have been completed, how much code has been produced, whether people are "busy enough" or not, or how far through the plan we are. Many will argue that those metrics were already the wrong ones, and I wouldn't necessarily argue with you.

But what is clear to me is that outcomes for the customer become even more important when a lot of the traditional measures become even less relevant. If vastly more output can be generated, those metrics around "volume of stuff created" become considerably less informative.

Instead, leaders will have to become much better at asking different questions. Rather than concentrating on how much has been produced or how busy everybody appears to be, we need to be much clearer about the outcome we were trying to achieve, what evidence we have that we're achieving it, what we've learned along the way and which of our assumptions have turned out to be wrong.

That moves leadership further away from monitoring activity and towards monitoring outcomes. Which, arguably, is where good leadership should have been all along.

The software team itself may change

Much of the discussion about AI and software development eventually arrives at the same question: "Will we need fewer developers?"

Maybe, but I don't think that's the most interesting question we could ask. A much more interesting one is what capabilities need to exist within an AI-augmented software team?

Today we divide software delivery into roles partly because different skills are required, but also because there is simply too much work for one person to do. Product management, business analysis, architecture, software engineering and testing are all, to a greater or lesser degree, separate roles performed by separate people. So we build processes, meetings and governance structures to coordinate all of those people.

Perhaps the shape of the team changes altogether.

But what happens when AI can perform meaningful parts of the analysis, implementation, testing, research, documentation and operational work itself?

Perhaps the seven-person software team doesn't simply become a four-person version of exactly the same team. Perhaps the shape of the team changes altogether.

The boundaries between disciplines could become softer, or disappear altogether. Smaller groups of highly capable people might become responsible for much larger portions of the end-to-end outcome, using AI to provide capabilities that previously required additional people, roles or hand-offs.

That might mean fewer coordination mechanisms, fewer hand-offs and maybe even fewer meetings. But it also requires much greater clarity about ownership and intent, which brings us full circle to the question of leadership in the world of AI augmentation.

Leadership influence might change too

There is another aspect of this that I find interesting. A lot of leadership influence today happens synchronously. Things get "escalated" to us for a decision because teams lack the confidence, experience or context to make those decisions themselves.

But increasingly capable AI systems can consume organisational context continuously, potentially removing some of the need for those escalations.

Instead of being the person who must constantly be present to provide direction, maybe it will become increasingly important that leaders create a high-quality context in which people and AI can make good decisions without constantly referring upwards. Better direction setting, in other words.

Good delegation is much richer

The more I think about this, the more parallels I see between that and good delegation. Poor delegation is simply "go and do this". Good delegation is much richer. It's about explaining the outcome we're trying to achieve and why it matters, being clear about the things someone can decide for themselves and the things they can't, and making sure they understand what good looks like and when they need to come back to you.

That is good leadership of people, and with AI augmentation it becomes even more essential. In fact, I'd argue you can't effectively use AI without it.

All of that makes me wonder whether leaders who are already comfortable leading capable, autonomous people will find the transition to AI-augmented organisations easier than leaders whose management style relies on supervising activity.

Micromanagement doesn't scale particularly well with humans, and it's going to scale even worse when your team includes machines capable of producing work at extraordinary speed.

Which is where I arrive at the title of this article.

I think poor leaders who muddle through with micromanagement and the support of a good team today might get found out tomorrow when their success depends on good clarity of direction and outcomes.

Perhaps, then, coming full circle, AI isn't going to require an entirely new model of leadership. Perhaps it's simply going to make bad leadership much harder to hide and considerably more expensive, and already good leaders will continue to flourish?

I'd love to know what you think.

About me

I write and share practical ideas about technology, leadership and transformation through The Common Sense CTO website and YouTube channel.

Through Stepthinking, I work with organisations that need experienced technology leadership, whether that's fractional or interim CTO support, technology strategy, transformation, operating model design or independent advice.

If any of that sounds useful, feel free to get in touch.