I have recently learned to code with words.
Why? Because I did not want to erase entire forests to print learning resources for my leadership courses, and the electronic copies felt clunky.
So, I braved a tentative step into the world of coding. Not the writing of HTML, CSS or JavaScript of course. That remains foreign to me.
More precisely, I learned to develop software through natural-language programming with AI assistance. In other words, I would tell the AI what I needed, and it would write the code for me. Simple.
Or so I thought.
Learning to work with AI to produce meaningful, functional and genuinely useful digital tools was much harder than I imagined.
My first tool was for the Eight Dimensions of Strategic Rigour: turning 42 pages of print into an interactive web-based resource where technology would serve precision thinking rather than dictate it.
After only a few attempts I realised that I was making way too many assumptions about how the AI would interpret my instructions. By the end of the first day, I understood that to get the outcomes I wanted, I would need greater clarity of intent, precision in communication, relentless attention to execution and constant verification of outcomes followed by adjustments and tweaks. This would be the only way to ensure I stayed true to my purpose.
In other words, to build the tool I needed for strategic rigour – I needed strategic rigour.
The irony was not lost on me. Here I was developing digital tools to support the very leadership thinking I help leaders bring into focus, only to discover that the process of creating those tools demanded the very same discipline.
With this realisation in mind and knowing exactly what the resource needed to do, I pushed forward with renewed enthusiasm. I knew how leaders should experience the tool. I knew what the content meant and why particular questions, sequences and distinctions mattered.
But with language-based software development, knowing what you want and communicating it with sufficient precision for someone, or something, else to build it are two different things.
And while AI is remarkably capable of filling in gaps, it does not necessarily fill them in the way you intended. An instruction that seems perfectly clear can produce something quite different to what was envisioned. An instruction for a seemingly minor adjustment can subtly change meaning. A request to simplify can remove a distinction that was deliberately included.
The leadership parallel is clear: Clarity is not established by what we believe we have communicated. It is established by what has been sufficiently understood to enable appropriate action.
When working with AI, that understanding cannot simply be assumed because the response sounds intelligent. AI can produce an exceptionally articulate explanation for a proposed change. It can sound confident, thoughtful and persuasive.
And it can still be wrong.
A polished explanation of a strategy is not evidence that it will be sound in practice, and it certainly is not a guarantee of outcome.
But perhaps the most important learning was about my very human self.
As the work progressed into the second night, the AI continued producing suggestions, explanations and revisions at remarkable speed, yet I was exhausted. At this point my eyes were bleary and I was struggling to focus. But with the digital tool almost finished and so many other things still on my to-do list the next day, I was determined to get it done.
And that was a recipe for trouble. I started to accept an AI recommendation here and there without fully examining the implications. The reasoning sounded highly logical – and at close to midnight that was good enough for me.
But when I checked the end result, the tool had drifted. It still looked professional yet some of the deep intent and precision had quietly drained away through fast thinking, a lack of verification and quick decisions based on assumptions.
That was unsettling.
Not simply because the technology had made mistakes, but because I could see the risk of abdicating thinking and judgement to AI.
On a small scale, my tiredness had briefly made me intellectually sloppy.
Luckily, the consequences were equally small and easily corrected with a few more hours of work. But it did make me wonder about the cost of such intellectual slippage when the stakes are higher.
Much of the discussion about artificial intelligence focuses on what the technology might eventually become capable of doing to us – but I wonder what, in over-relying on AI, we might do to ourselves.
There is a fundamental difference between an AI that helps us think and an AI that thinks on our behalf.
And there are parallels to leadership here too.
In educational leadership, we frequently operate under conditions that can be conducive to cognitive shortcuts: urgent and competing priorities, incomplete information, limited time, persuasive advice, the pressure to act and sometimes sheer exhaustion.
So perhaps with both AI and strategic leadership, the greatest danger may not be that we consciously decide to surrender our judgement and intellectual rigour. It may be that under pressure, we gradually stop noticing when we have done so.
