The Real Skill for the AI Age Is Unlearning

When I was at university, I had a friend who was a master bodger.

When something broke, he didn’t replace it, he engineered around it.

The front door of his flat only opened if you pulled it toward you and jiggled the key at just the right angle. The door of his kitchen cupboard hung from a hinge fashioned out of duct-tape. His sink didn’t have a plug, so he repurposed a jam-jar lid that was almost the same size.

Any sane person saw this as madness, but he was proud of his makeshift solutions. Each of them was a small victory of ingenuity over inconvenience. But each bodge resulted in more effort and a worse experience.

And this is what I’m reminded of when I look at companies preparing to implement AI. They’re hoping their Copilot implementation will magically cut overheads and replace human effort, while overlooking the decades of corporate cludges that have led to a tapestry of badly patched processes.

And AI tools are more likely to amplify these dysfunctions than magically erase them.

Learning’s essential - but it’s only half the job

I make my living as an educator, so let me start by making it clear that learning matters.

Especially in a world that’s shapeshifting faster than an X-Men character.

However, the traditional approach to learning focuses only on addition. It’s about picking up new tools, new knowledge, new skills. But that’s only half the equation. The other half is subtraction. It’s about letting go of what no longer serves us and the assumptions that hold us back.

To address this, I often run an “assumption-shattering” exercise with teams. I get them to list all the beliefs embedded in a process and then challenge them one by one. It often leads to some uncomfortable realisations - that a lot of what people treat as gospel hasn’t been questioned in years.

Unlearning isn’t anti-learning; it’s the natural counterpart to any educational effort. It's like a conscious version of neural pruning, where we get rid of what's not valuable to us to help keep our thinking nimble and efficient.

Without it, we just keep piling new ideas on top of outdated ones - like installing solar panels on a collapsing roof.

The three stages of AI adoption

Education doesn’t stand alone. It’s an essential part of an organisation’s AI journey.

There’s a model I use to help organisations work out what stage of maturity they’re at on their AI journey. It’s a super-simple, three-step pyramid.

Stage 1: Reduce

This is where every organisation starts - and where most of them stay. It’s all about using AI to reduce costs and time: do more, faster, cheaper. Automate the reports. Draft the emails. Summarise the meetings.

There’s nothing wrong with this. Every company needs to earn back some time. But progress stalls when “faster” becomes the only goal.

What needs to be unlearned here:

• The idea that productivity equals progress.
• The reflex to measure success only in hours saved or costs cut.
• The habit of treating technology as a bolt-on instead of a catalyst for change.

Unlearning at this stage means widening the definition of value from output to outcome, from efficiency to effectiveness. It’s about seeing AI not as a robot intern but as the start of a deeper rethink.

Stage 2: Augment

A minority of organisations move on to Stage 2. This is where AI is used to enhance human capability: to stretch thinking, raise quality, and open new possibilities.

But many leaders find this stage uncomfortable, because it forces them to let go of the old power dynamics. The expert who once had all the answers now needs to ask better questions.

What needs to be unlearned here:

• The belief that authority comes from knowing rather than learning.
• The instinct to protect turf instead of inviting collaboration.
• The assumption that AI’s role is to replace, not to expand, human contribution.

Unlearning at this stage means creating a culture of curiosity and humility. A culture that rewards experimentation and accepts that progress looks messy before it looks clever.

Stage 3: Discover

A few companies are sniffing around Stage 3, but I’ve yet to see one fully embrace it. This is the frontier: using AI not just to do work differently but to reimagine what work even is.

It requires the most unlearning of all, because there’s no familiar playbook to lean on.

What needs to be unlearned here:

• The reliance on past models of success.
• The fear of ambiguity and exploration.
• The separation between “business strategy” and “creative thinking.”

Unlearning at this stage means letting go of certainty. It means embracing discovery as a habit - making space for new questions, new forms of value, and new ways of organising human and machine intelligence together.

I can’t wait to work with clients on Stage 3 projects, because that’s where it gets truly exciting.

However, with so many companies stalled at Stage 1 - held back by old assumptions and compounded dysfunctions - I suspect that’s where most of my work will be for the foreseeable future.

What happens when we don’t unlearn

AI doesn’t work as a sticky plaster for dysfunctional systems. Instead, it scales them. Poor decisions, petty politics, bias and inefficiency all become faster, slicker and more problematic.

When organisations fail to unlearn the old ways - old systems, old silos, old assumptions - the repercussions are real and often show up on your balance sheet.

More than six in ten firms still rely on legacy software frameworks in 2025, even while admitting those systems are hard to integrate, insecure and expensive to maintain.  Apparently the average cost of keeping one outdated system operational is around $40,000 a year — before you even factor in the knock-on effects.

Then there are data silos: when information is trapped in departmental systems, unable to be shared, unable to feed insight across business functions, the organisation becomes fragmented. One IDC analysis places the global cost of these silos at $3.1 trillion annually.  Another finds that siloed data increases operational costs substantially by duplicating work, slowing decisions and undermining collaboration.

Yet most companies ignore this stuff while they trim costs elsewhere.

Add to that the technical debt of maintaining aging infrastructure: one survey found enterprises losing hundreds of millions of dollars each year simply because their transformation projects were hampered by outdated systems.

So without unlearning the old - the legacy systems, siloed databases, old habits, unchallenged assumptions - even the smartest organisations end up reinforcing yesterday’s logic with today’s technology.

Unlearning is what turns learning into progress.

Leadership and the courage to let go

Unlearning can’t happen by memo. It has to be modelled, resourced, and rewarded.

We need to build unlearning into the system: making space for reflection, celebrating people who retire outdated processes instead of protecting them.

This isn’t about reckless change. It’s about clearing the mental clutter that slows everything down. Out-of-date knowledge is like a bungee rope holding you in the past. The harder you pull forward, the tighter it snaps back.

Don't unlearn this

So here’s my real call to action:

Don’t myopically focus your efforts on learning the new when the old is holding you back.

I'm not saying it's easy. It's not. It often involves kicking away a comfortable crutch and walking into an unknown swamp. But you don’t need to do it alone.

Unlearning isn’t a rejection of knowledge. It’s how we make room for what’s next.

Dave Birss is Co-Founder of The Gen AI Academy - a collection of over 30 AI experts offering practical training and advice for companies all over the world. If you haven't taken his LinkedIn Learning courses, you really should.