What is AI thinking?
And do you need it?
Your organisation’s ability to win with AI will not be determined by your spend, technology, capability or talent, but how you think. AI means that many of us need to learn how to think all over again.
Sounds simple, right? And it is, sort of, at least for those that can recognise a new way of thinking is required for the AI era. In this article I take a deep dive into the need for a new way of thinking about AI; not purely as a technology, but as way of working and doing business. This is for you if you recognise the structures you put in place to help you deliver defined software that behaves in an understood way, for a defined group of people, in a way that you can control - might not be quite right, now that AI is here. Let’s get into it.
Mostly, organisations have now figured out how to deliver digital transformation. They define what to build, plan how to build it, measure and govern it along the way. Lean, agile, design thinking approaches accelerate product market fit.
This model works brilliantly for software that does the same thing every time.
The frameworks are known, accepted, understood, adopted and adapted. Deeply entrenched muscle memory has been established from honing software delivery over decades. Yet a problem emerges when the de facto digital ways of working are being used to realise the next wave of emerging technology opportunity. When the same thinking is being applied to AI.
Move fast and… deliver AI
Organisations at different stages of AI maturity are implementing AI with urgency. Often early AI hires are technical: engineers, builders, busily getting stuff done. But what, exactly? And why?
Every board wants to see the AI roadmap with the right blend of operational, product and strategic AI. But planning with fixed roadmaps and measuring in the metrics they built for deterministic systems is problematic. It is a structure built for something else, and so it’s no surprise to me that AI isn’t delivering for organisations.
Build, Test, Use, Repeat
In a deterministic world, software does what it’s told; every time. Once your software is built, tested, and out in the wild it will just keep going unless something changes. And to a greater extent, we trust that what we have created, will persist, will continue to do its job as we intended. As product leaders, portfolio owners, C-Suite teams responsible for software, in the deterministic world we can sleep easy at night.
The new world order
AI systems are not deterministic. They produce probabilistic outputs: ‘more or less likely’, not ‘right or wrong.’ They learn from use, they change over time, in ways you didn’t design.
And not just model drift, AI systems create second-order effects that you hadn’t envisaged when you started. The platforms you use to power your AI products change rapidly, the regulation plays catchup, governments panic. Risks you used to imagine were super low like models being banned by governments have happened in recent weeks. Customers are using AI too, and becoming fluent in the conversation around data, ownership, governance, second order consequences. All our roles are changing and it’s getting complicated.
A new way of thinking
How can you better anticipate unintended consequences? How can you anticipate entirely new forms of risk? How can you design systems that intentionally learn and evolve, in ways you might not be able to anticipate? These are questions that cannot be solved with existing frameworks; frameworks we’ve used to deliver, even the most groundbreaking software and services for the last two decades. Agile only gets you to delivery faster. Lean only gets you better product market fit. Design thinking helps identify problems to fix. Exponential thinking helps you think really really big, but doesn’t help you deliver, integrate, adopt.
We need a new way of thinking as we ideate, plan, build, evaluate, iterate, moderate and even use AI; we need AI thinking.
What is AI thinking?
First let’s start with what it’s not. It’s not AI literacy (which is knowing about AI tools and terminology). It’s also not AI fluency (which is the ability to use and the experience of using AI tools).
For me, AI thinking is the organisational and operational style that determine whether your organisation can win with AI. It might borrow from exponential thinking in requiring you think about compounding, it might borrow from lean, in that it requires you to iterate and test, it might require you to be agile, in that you don’t know all the answers at the start. It requires all of that and more. For me, AI thinking has six characteristics.
AI thinking is probabilistic
The AI era demands ‘comfort with uncertainty’ in a way no previous tech era does. Most of us (tech leaders) were trained to deliver certainty: confident forecasts, clear recommendations, strong decisions. I pride myself in my abilities in all these areas; skills honed by decades of experience, thoughtful reflection, practice, learning from the best.
AI thinking requires a different approach. We now need the ability to reason in likelihoods not certainties, we have to make bets with partial information, build in checkpoints later, rather than do all our diligence up front. We need to communicate uncertainty to boards and teams without it reading as indecision. We need to help them be ok with it too.
This is a specific and pretty enormous cognitive shift, and is not just a “tolerance for ambiguity”. It means understanding that “I don’t know exactly, but here’s how we should position for a range of likely outcomes” is more useful than false certainty.
We have to be okay with “we don’t know it all”.
AI thinking is anticipatory
Imagine suddenly being able to spot patterns in the data you’ve sat on for decades, allowing you to provide entirely new products to your customers. Imagine preempting, rather than responding to your customers’ problems. When you start, you know you’ll be able to do new things, but you can’t exactly plan for them.
AI thinking requires this ‘forward orientation’. To even imagine outcomes you need to track the signals that tell you what’s coming before it’s obvious, and think through the implications early enough to do something about them.
This isn’t forecasting, but following the frontier AI and innovators’ capability curve closely enough to avoid being caught off guard by changes that were visible to those paying closer attention.
I like the analogy of skating to where the puck is headed.
AI thinking is iterative
Wrong bets teach you something; right answers teaches you nothing. Iterative in the AI context, means expecting to work via learning loops over structures and processes designed to engineer perfection.
It’s about being ok with the fact you are learning.
It is also about being able to learn from mistakes and knowing when to stop. The ability to kill something quickly when it isn’t working without the sunk cost mentality.
Speed without the discipline of learning when to flip the kill switch is how organisations end up with a portfolio of AI pilots that deliver nothing and create a lot of noise.
Experience and ego may have served us well in the more predictable past, but now we have to be prepared to iterate.
AI thinking is compounding
AI can help us build things that appreciate over time rather than just solve today’s problems. In fact, that’s how we should see it; and yet, so many organisations are still held up with implementing AI ‘at the edges’.
Compounding AI thinking goes beyond adding features to improve metrics; it’s about building data assets that get more valuable with every interaction, building the capability that makes the next capability cheaper to build, making the decisions now that create optionality later. The question we should ask is not “does this work?” It’s “does this compound?”
Organisations that will be structurally ahead in two to three years’ time are building things that compound now.
AI thinking is translational
The AI era requires leaders to move fluently between technical depth and commercial clarity; to understand what is actually being built well enough to make good strategic decisions about it, and to translate that into language that makes sense to a board, a leadership team, an investor or a customer.
If there’s a gap between the people who understand what AI systems are doing and the people who need to make decisions about them, failure follows.
In the old world of deterministic systems, commercial decision makers interacted less with the builders. Some level of technical knowledge was required, but a deep understanding was not. AI leaders do not all need to be technical but the conversations between specialisms must be deeper. Understanding must be deeper both ways.
It’s about talking more and reaching a shared understanding.
AI thinking is generative
The final characteristic is the one that distinguishes AI leaders from AI implementers. Generative thinkers generate; they build capability in others, not just outputs for themselves. They create the conditions in which the people around them think more clearly, more strategically, more practically about AI.
AI capability that lives in one person or one team is fragile. It doesn’t scale, it doesn’t survive staff churn and it doesn’t change how the organisation actually operates. Generative AI thinking builds something much more durable: an organisation that can leverage AI, not just use it.
Objections?
There are always objections: “we don’t have time to rethink how we think… we need to move now…”
I understand the urgency; the cost of moving slowly in the AI era is high. But urgency without the right way of thinking is frivolous activity that doesn’t compound into advantage, will create the sort of waste we see with ‘failed AI pilots’, and will ultimately lose you places in the AI race.
We don’t have to learn everything from scratch, but we need the fleet of foot to build new mental models, new responses, new lines of enquiry, to learn to learn again.
Rebel Futures helps business and technology leaders navigate AI-driven disruption. A future piece in this series looks at what an AI-thinking organisation actually looks like, and how to build one.