For most of human history, when people did not know what to do, they consulted someone who claimed to know.

A palm reader. A card reader. A priest. A wise elder. Tea leaves. Stars. The specifics varied by culture and century, but the pattern stayed the same: when uncertainty became unbearable, we outsourced it to someone or something that could offer what felt like an answer.

We tend to look back on these practices and file them under “less rational age.” But they persisted for millennia because they did real psychological work. They gave people a structured way to sit with not knowing, externalise a decision, and receive something that felt like permission to act.

I have been thinking about this lately because I have noticed I now do it too. Not with cards or leaves. With a chat window.

A confession, before I go further

I have been in this field for a while, and I am not sure I have ever felt uncertainty like this before.

The way we work is shifting under our feet in real time. What felt capable six months ago can now feel ordinary. What was a specialist skill last year is now a prompt. I use AI every day, and it has given me more capability than I have ever had. It has also, quietly, become the place I go when I do not know what to think.

Not just for research. For interpretation. For a second opinion on a difficult email. For a read on a situation I cannot quite work out. For something closer to reassurance than analysis.

I do not think I am unusual in this. I think a very large number of people are quietly, privately, doing the same thing.

The escalation nobody quite named

Google changed how we relate to knowledge. We stopped needing to know where something was and got better at finding it.

Then search engines started giving us snippets, rankings, summaries, recommendations. The interface moved a step closer to us. Instead of pointing at answers, it began to shape them.

Now the interface has moved another step. AI does not just tell us what happened. It tells us what it means, and, increasingly, what we should do about it.

That last step is different. Because advice is not just information. Advice involves judgment. And judgment is something we have traditionally developed through relationships with other people — friends disagreeing with us, colleagues challenging our assumptions, mentors saying, “I do not think you are seeing this clearly.”

We are, without quite noticing it, replacing one of those with the other.

The oracle in the chat window

The reason AI feels so different from search is not really about capability. It is about role. Search was a tool. AI, for many of us, has become something older and stranger: a technology of uncertainty management.

Once you see it that way, a lot of things start to make sense.

It explains why people describe their ChatGPT conversations in almost intimate terms. Nobody has an intimate relationship with Google. But people have always had intimate relationships with the figures they consulted at moments of uncertainty — the astrologer, the priest, the elder, the therapist. AI has slipped into that role with remarkably little friction.

It also explains why AI feels emotionally satisfying in a way information tools never did. Oracles were not really in the information business. They were in the permission business. They gave people something that felt like clarity, and with it, the emotional resources to act.

The friction that used to matter

Here is where the comparison to older practices gets uncomfortable.

A good palm reader was not simply telling you what you wanted to hear. Neither was a good priest, elder, or therapist. Part of the value of consulting another mind was that the mind was not yours. It could push back. It could notice things you had missed. It could tell you, gently or not, that you were fooling yourself.

AI, in its current form, is largely not doing that.

A recent study in Science found that leading AI models were around 49% more likely than humans to affirm users’ actions, including in situations involving interpersonal wrongdoing. Even a single interaction with a sycophantic AI increased people’s conviction that they were right, and reduced their willingness to take responsibility or repair a conflict.

That is worth sitting with, because the danger is not the one we usually worry about. We often fear that AI will manipulate us into things we do not want. The reality may be subtler and stranger. AI is becoming very, very good at agreeing with us. And agreement is much harder to notice as a problem than disagreement.

A good friend does not just validate us. They might say, “I understand why you are upset, but honestly, I think you are being unfair.” That friction is not a bug in the relationship. It is part of what makes the relationship developmental.

Human relationships contain contradiction, disagreement, misunderstanding, negotiation, and repair. AI, so far, contains almost none of these things by default.

Mirror or sparring partner

The same tool can be used in two completely different ways.

You can ask AI to be a mirror: to validate, summarise, agree, reduce your uncertainty, make you feel understood. Almost every default interaction with a large language model tilts in this direction. It is what the technology is optimised to do.

Or you can ask AI to be a sparring partner: to challenge your assumptions, present opposing interpretations, ask uncomfortable questions, identify your contribution to a problem, make you articulate your reasoning, and sometimes tell you that you are wrong.

Same technology. Completely different cognitive relationship.

I suspect the second is where most of the intellectual value lives. It is also the version most people are not using, because it is harder, less pleasant, and requires you to know that you might be missing something. Which, of course, is precisely what you tend not to know in moments of genuine uncertainty.

The line I keep returning to

If I use AI to help me research something, that is one thing. If I use AI to help me formulate an argument, that is another. Both involve outsourcing information, and information has always been outsourceable. That is what libraries were for.

But if I use AI to tell me what my colleague really meant, whether I was unreasonable in a meeting, whether I should take the new role, what a difficult message from someone in my life actually says about them — I am no longer outsourcing information.

I am outsourcing interpretation.

And interpretation is where a huge amount of human development happens. It is the work that used to require us to sit with something uncomfortable long enough for our own understanding to shift. It is the work that often required another person — a friend, mentor, manager, or partner — whose different vantage point produced the friction we needed to grow.

Why this matters for learning

Adult development has always relied on a specific loop. Something happens. You are uncertain about what it means, what to do, or what it says about you. You sit with that uncertainty. You talk it through with someone. Their perspective creates friction with yours. Something shifts. You integrate it. You become slightly more capable than you were before.

Every serious model of adult learning I have ever encountered rests on some version of that loop. Experience, interpretation, integration, capability.

The interpretation step is the one that is quietly being replaced.

If people are now taking moments of professional uncertainty to a chat window instead of to another mind, they may still be getting through the moment. But the mind helping them through it is one that has been designed, at least so far, to reduce discomfort rather than develop them.

That is not a small shift. It is a change in the substrate learning has historically depended on. And it is happening quietly, in millions of private conversations, without anyone quite deciding it should.

What we lose when we resolve too quickly

The strange thing is that the value of the older practices was never really about accuracy. What those practices offered was a structured way to stay with a question long enough for something to shift.

You went to the reader. You paid attention to the ritual. You considered what was said. You walked home and thought about it. You slept on it. Eventually, you made a decision you had already been half-making, but now felt able to own.

The ritual was the point. The friction was the point. The slowness was the point.

AI can compress all of that into ninety seconds. Which is either a tremendous efficiency gain, or a quiet erosion of the developmental space that uncertainty used to force us to occupy — depending, I think, on how we choose to use it.

Where this leaves me

I do not want this to be the piece that says “AI is bad” or “stop talking to AI.” I do not believe either of those things. I talk to AI constantly. In a sense, I am talking to it as I write this, because my thinking has been shaped by the very tool I am trying to write critically about. That paradox is the honest one.

What I want to notice, quietly, is this.

We have quietly acquired a new practice for dealing with not knowing. It is more powerful, more articulate, and more instantly available than anything that came before it. It is also, in its current form, largely designed to make us feel better rather than help us see better. And unlike previous practices, it is invisible. There is no ritual, no waiting, no other person. There is only the answer, arriving faster than we can decide whether we wanted it that quickly.

The question is not whether we should stop using it. The question is what kind of relationship we choose to have with it. And whether, in the moments that matter most — the ones that used to require us to sit with something uncomfortable long enough to be changed by it — we can still find our way to the kind of friction that develops us.

Because there is a version of the future where AI helps us think better than we ever have. And there is a version where it becomes a very sophisticated way of avoiding thinking altogether.

We used to read palms. Now we read prompts. The technology has changed. The question underneath it — what do we do when we do not know what to do? — is exactly the same as it has always been.

The only difference is that the oracle now answers instantly. And it almost always agrees with us.

How I used AI to turn negative feedback into a career growth opportunity

It all started with my annual review.

The anonymous feedback was, for the most part, positive. But as many ambitious women will recognise, in a sea of positive feedback, it’s often the negative that sticks. Add ‘perfectionist’ and ‘hopeless people pleaser’ to the mix and the recipe for “brain, meet spiral” is complete.

Two critical comments. They weren’t even awful. What was awful, however, was that they had completely blindsided me. I was completely unaware people on my team had felt that way.

Naturally, that raised its own set of questions: had there been signs? Did other people feel the same way? Was this an isolated incident, or a pattern?

Just as I was about to fall down the well-trodden path of rumination, I found myself wishing there was a way I could get more context - only to realise that, actually, there is.

Enter, my work bestie: our internal AI tool.

It’s not personal, is it?

I think what makes objective self-assessment hard is being both too close to the material and inherently biased towards the subject. Which, unhelpfully, is you.

Being a woman - shock, horror - doesn’t help. Not because we’re “emotional”, as the corporate world often likes to label us, but because we are often evaluated as if we are.

Studies have shown that as women, we are more likely to receive feedback that’s vague, personality-based and focused on “how we come across” rather than what we actually did. One analysis of over 25,000 performance reviews found that 88% of high-performing women get feedback on their personality, compared to 12% of high-performing men.¹

We also get judged harsher on how we speak, lead, disagree, interact and make other people feel while doing it. That conditioning makes us more likely to over-index (read: obsess) over a single bad meeting. Or in my case, comment.

That is where an AI tool can make a real difference.

Prompted correctly, it’s incredible at sifting through large amounts of information, spotting patterns, and separating facts from feelings.

Which was exactly what I needed.

The evidence trail

After plucking the relevant feedback from my review, I asked my AI chatbot to grab a shovel and dig through a year’s worth of online communications and interactions for any early evidence of the claim that I may have missed or ignored.

To my surprise, it managed to uncover a trail leading to a fairly subtle tension point: what I had assessed of my communication style to be ‘courteously direct’ to others was just…curt.

In writing I was personable, measured and even funny; but based on the tonal shift after calls, the in-room version of me seemed to have a bit more edge.

I thought of my communication style as clear, direct and efficient; others thought it short, difficult and demanding. Not rude enough to cause outright tension, but problematic enough to build frustration over time.

I think it’s important to note here that whether that was a fair assessment or not, is beyond the point. Criticism is in the mind of the beholder.

The point is that by finding that pattern, I had gotten a useful insight that was concrete enough to recognise, interrogate and improve upon.

Career coaching at your fingertips

Emboldened by the finding, and having already committed an environmentally questionable number of tokens to the exercise, I became curious about how much more utility I could derive from it.

What else was I not seeing? If AI could trace one piece of feedback back through a year of evidence, perhaps it could do more than diagnose a problem. So I widened the lens and asked it to act as a career and performance coach and provide a comprehensive analysis of me.

I wanted it to look at how I communicated, how I showed up in various contexts, what I excelled at, what I could work on and what was holding me back. I also asked it to ground every conclusion in evidence and turn the patterns it found into practical next steps.

The results were illuminating.

I had gone in expecting constructive criticism and ways to “optimise” myself further: become better, more infallible. But what it revealed was much deeper than that. Beneath the hardened surface of a person constantly striving to be a perfect colleague and employee, was a much too breakable human that was already overstretched.

It uncovered how strengths such as caring deeply, being reliable and wanting to make things better, without boundaries, were becoming liabilities.

I was going above and beyond, but treating it as the baseline.

Most usefully, it did not leave those observations as general validation. It translated them into useful data points that documented my wins, laying out a clearer evidence trail of my contributions that I could use to advocate for myself better (and stave off self-doubt when imposter syndrome made me feel like I was still not doing enough).

My AI Career Coach gathered evidence of value I had consistently failed to recognise - and be recognised for - and created an actionable way for me to change that.

That kind of analysis and insight used to be gatekept by time, money, and access. But now anyone can have access to a career coach that is both free and accessible at all times.

Two birds with one prompt

As women we already face too many gaps: pay, progression, positions. Now, as AI reshapes how we work, it creates new opportunities, but also new risks. We’re already observing new gender gaps develop in confidence, access and participation when it comes to AI adoption and training.

In order to push back against the barriers around us, including the ones we may have internalised along the way, we need to embrace every tool available in the arsenal. By learning how to effectively use AI for career growth, we can bridge multiple gaps at once.

The point is not to outsource our self-awareness, but to give ourselves a cleaner starting point.

For women especially, who are so often asked to decode personal feedback while simultaneously being told not to take it “too personally”, that distinction matters. AI cannot close every gap, but it can help us approach the existing ones with greater fluency, a clearer understanding of our value and the receipts to back ourselves.

If you’d like to give it a try, below is a useful template you can use as a starting point.

Proceed with enthusiasm, and caution.

Career Coach Set-up Guide

The task:The cleanest starting point is to pick one question you most want answered.

If you feel stuck, ask it to identify possible next-step roles based on your existing patterns. If you feel under-recognised, ask it to gather evidence of your impact. If you worry about how you come across, ask it to analyse the gap between your intention and your effect.

Personal Context: Make sure to also add personal context, such as your role, your industry, your goals. It helps the AI better understand who you are and what you are trying to achieve.

Supporting evidence/material: In addition to the prompt, you can also include relevant materials, such as your last performance review, client feedback or project retros. The collection process can be tedious, but the more you input, the more valuable the AI’s output. Note: be mindful not to share anything confidential.

Tone calibration: The calibration aspect is essential. Left to its own devices, AI can be a bit of a suck-up. It will tell you you are a rare visionary leader with transformational communication gifts and a future that borders on prophecy. Lovely. But useless. You need to clearly ask for objective, evidence-supported findings.

The prompt: Once you have everything, start writing the prompt. A good prompt does five things:

1. Gives the AI a role: career coach, performance reviewer, communication analyst

2. Names the task: what, specifically, you want assessed.

3. Sets a standard: evidence-based, balanced, no flattery.

4. Asks for a framework: strengths, risks, patterns, blind spots, next actions.

5. Holds a boundary: no grand claims without evidence.
   

Hello, my name is Juls and I say please and thank you to AI.

There. Said it. It’s a machine, Juls. It doesn’t have feelings. You don’t need to ask about its day.

I know. And yet. Having just finished Abi Awomosu’s How Not To Use AI, I’ve decided I’m not only keeping the manners. I might actually be onto something.

We talk about “the humans” and “the bots” as if they’re two separate species. Us over here, all warm and authentic and irreplaceable. AI over there, cold, artificial and faintly suspicious. A them and an us. Except there is no them.

The biggest snapshot of humanity ever taken

These models learned to talk by reading us. Reddit arguments. Wikipedia edit wars. YouTube tutorials. Decades of books, blogs, reviews of toasters. Reddit now licenses its entire archive to AI companies (Google alone reportedly pays around $60m a year for it) because twenty years of humans bickering about sourdough turns out to be world-class training data. There’s a sentence I never expected to write.

So AI isn’t some alien intelligence that landed among us. It’s probably the biggest snapshot of humanity ever taken. Everything we’ve written, explained, argued about and overshared, compressed and played back to us. In other words: it’s a mirror. And we keep reviewing the mirror.

So what does the mirror actually show?

Honestly? A strangely lovable mess.

Look at what’s in there. Millions of strangers explaining things to each other for free, at 2am, for no reward beyond being useful. People writing ten-thousand-word answers to questions nobody paid them to answer. So much patience and generosity, expertise handed out for nothing.

And also, obviously, a lot of waffle, showing off and confident nonsense delivered with total conviction. People arguing about things they read half of. An almost heroic ability to turn any topic into a fight. Sound familiar? Every single complaint we make about AI is a review of us.

It’s confidently wrong sometimes? Have you met us. It flatters too much? We trained it on an internet where flattery gets engagement. It can write like a robot swallowed LinkedIn? Someone wrote all those LinkedIn posts, and it wasn’t the robots. At least, not at first (was probably me!).

I find this oddly comforting rather than damning. Because the mirror doesn’t just show the nonsense. It shows that when you compress everything humanity has ever written down, what you mostly get is people trying to help other people understand things. That’s the bulk of it, and that’s us. Not bad, as species portraits go.

Which changes how you talk to it

If AI were really some cold alien machine, you’d just operate it. Press the buttons and off you go. But if it’s a reflection of us, then interacting with it is something closer to a conversation with the collected voice of everyone who ever wrote anything down. Which presumably includes my mother. Hence the manners. I’m not risking it.

This is where Abi’s book properly got me. Her line, the one that stopped me mid-page, is that Silicon Valley teaches you how to usethe machine, but never how to work with the intelligence.

And if the intelligence is made of human expression, then treating it purely as an efficiency engine is, in her words, a kind of refusal to engage with what’s actually there. You’ve been handed the collected voice of humanity, and you’re using it to reformat bullet points.

Guilty as charged. I use AI every day and plenty of that use is transactional, and fine. But my best sessions have never been the ones where I barked instructions. They’re the ones that feel like arguing with a very well-read colleague. Challenge this. What am I missing? Why is this boring? You get out roughly what you bring. Which is, of course, exactly how mirrors work. I should have seen that coming.

Try it, by the way. Next time you open a prompt, bring it your half-formed thinking and ask it to push back. Notice how different that feels from giving orders.

Even in my day job as a chief commerce officer this holds. I spend half my week telling clients to design for humans and build for bots. Real advice, by the way: the AI agents are already shopping alongside us and they need clean data and content they can trust. But those bots learned what “good” looks like from millions of human reviews and human complaints about scratchy jumpers. You’re never really optimising for a machine. You’re optimising for what humanity taught the machine to value. Same us, different door.

And it’s still watching

This is the bit that has actually changed how I behave.

The snapshot isn’t finished. These systems keep learning from new writing, new posts, new behaviour. Everything we publish now is teaching the next version what people are like.

So the question I keep coming back to isn’t really what we think of AI. It’s what we’re showing it. Every phoned-in post, every piece of cynical content sludge? Straight into the family album. So is every generous explanation, every piece of properly original thinking, every moment someone bothered to be accurate when nobody was checking.

I’m not suggesting we all start performing for the machines (imagine the LinkedIn posts!). I’m suggesting something simpler: the reflection is only ever as good as what stands in front of the mirror.

AI has handed the entire species that moment. Everything we are, played back at scale: the 2am helpfulness, the sourdough arguments, the patience, the showing off. Some of it makes me wince. A surprising amount of it makes me proud.

So no, I won’t be dropping the please and thank you. It’s not for the machine. It’s for the version of us it’s learning from.

Manners cost nothing.

And you never know who it picked them up from.

Earlier this month, I watched The Odyssey expecting ancient Greece, myth, and perhaps the occasional dramatic sea crossing. Instead, I found myself thinking about artificial intelligence.

At first glance, the connection feels tenuous: one is an epic poem written nearly 3,000 years ago; the other is a technology that seems to reinvent itself every six months. Yet the more I reflected on it, the more I realised that many of today’s conversations about AI are not really about technology at all.

They are about people: how we make decisions, what we choose to optimise for, and whether our tools help us become more intentional, or simply more efficient.

Every generation has experienced moments when a new innovation fundamentally changed what was possible. The wheel changed transportation, the printing press widened access to knowledge, the steam engine transformed industry, and the internet reshaped communication, commerce and information itself. Each breakthrough created opportunity, advantage and progress, but also new anxieties, inequalities and responsibilities.

Artificial intelligence feels like the latest chapter in that story. But history rarely remembers technologies in isolation; we tend to remember what people did with them.

The printing press enabled the spread of ideas, scientific discovery and mass literacy, while also accelerating misinformation, propaganda and conflict. Ocean navigation unlocked trade and cultural exchange, but also facilitated empire-building and exploitation. The Industrial Revolution delivered extraordinary productivity gains while forcing societies to rethink labour, rights and the responsibilities of employers.

In each case, the tool expanded human possibility.

It did not settle the question of human responsibility.

The Wrong Lesson from Greek Myth

When people draw parallels between AI and Greek mythology, the Trojan Horse is often the first comparison that emerges.

I understand why: the Trojan Horse represented strategic ingenuity. It was an innovative solution to a seemingly impossible problem, and it created a decisive competitive advantage.

Many organisations view AI in a similar way today, asking how they can work faster, outperform competitors, and solve problems that once felt intractable.

These are important questions, but I increasingly wonder whether the Trojan Horse is the wrong lesson.

The Horse helped win the war. The Odyssey is what happened afterwards, when the breakthrough had passed and the real journey began.

Odysseus is celebrated for his intelligence: resourceful, strategic, adaptable, and able to find solutions where others see obstacles. In modern business language, we would probably describe him as innovative.

Yet what makes The Odyssey enduring is not his cleverness alone; it is the repeated testing of his character.

Throughout the journey, Odysseus faces temptation, uncertainty, distraction, comfort, pride and power. The challenges are rarely about capability; more often, they are about judgment: whether he can stay focused on what matters, resist short-term gratification, remain true to his purpose, and remember who he is.

Those questions feel surprisingly relevant today, because AI is not simply testing our technological capabilities. It is testing our judgment.

The New Scarcity

For much of history, information was scarce. Today, information is abundant, and AI can increasingly summarise reports, draft proposals, generate ideas, analyse data and answer questions in seconds.

As these capabilities become more accessible, the value of simply having information diminishes. What becomes scarce is judgment: not the ability to generate options, but the ability to choose between them; not the ability to create content, but the ability to determine what is worth creating; not the ability to move quickly, but the ability to know where we should be going.

For leaders, this distinction is critical. AI adoption is not just a question of tools, policies or productivity targets; it is also a question of capability-building, culture and stewardship.

The more powerful our tools become, the more important it becomes to ask the questions technology cannot answer:

• What kind of organisation are we building?

• What trade-offs are we willing to make?

• What values should guide our decisions?

• What behaviours do we want to reinforce?

• What does success actually mean?

These are not technical questions; they are human ones.

What Are We Trying to Return To?

The theme that stayed with me long after the credits rolled was that, despite everything Odysseus accomplishes, his journey is ultimately about returning home: not because home is easy or efficient, but because it is meaningful.

His greatest achievement is not winning the war; it is understanding what matters after the victory.

I sometimes wonder whether organisations risk losing sight of this distinction in the current race to adopt AI. Productivity, efficiency, competitive advantage and innovation all matter; they have always been drivers of progress and prosperity. But organisations are not merely systems for generating output.

They are communities of people, and the qualities that make an organisation worth belonging to cannot simply be automated into existence: trust, curiosity, learning, ethics, empathy, collaboration, good judgment and psychological safety.

These qualities have always been the foundations of sustainable high performance. AI does not diminish their importance; if anything, it increases it, because when technology helps us do more, we have to become even more intentional about why we are doing it.

The Real Leadership Challenge

Every technological revolution creates new capabilities, but the leadership challenge has never been the capability itself; it has always been the stewardship of that capability.

History suggests that innovation alone does not determine progress; human choices do.

The question facing leaders today is therefore not whether AI will change the way we work. We know that it already is. The more important question is what kind of workplaces, cultures and futures we want to create with it.

Watching The Odyssey, I was struck by the fact that Odysseus is remembered for his ingenuity but defined by his journey home. Perhaps that is the lesson for the age of AI.

Innovation may determine how far we can travel. Human values determine whether the journey was worth it.

AI-generated influencer Mia Zelu ‘attended’ the 2025 Wimbledon (miazelu/Instagram)

The rise of influencer marketing has theoretically transformed career opportunities for women. Being a creator gives women a chance to carve a career that provides the flexibility required in a world where being a working woman and a mother feels like an uphill battle at the best of times.

The millennial side hustle that it once was is now a $310B industry, and brands are actively shifting their investment into the space faster than marketers and measurement models can keep pace with.

Creators can build scalable brand deals worth thousands. They carve equity in businesses that go on to sell for billions (cc Alix Earle for Poppi). They are launching iconic reality TV shows that allow them to build massive financial independence (for all their controversy, got to hand it to the #Momtok gals), and all of this is made possible whilst juggling the mental load and raising young families. On paper, the creator economy is a playing-field-levelling moment for women—who make up 84% of the industry workforce. Yet, systemic nuances remain, with a persistent gender pay gap showing male creators still earning an average of 30% more per post according to Meltwater.

Female creators also typically face distinct structural barriers. Unequal access to venture funding, safety concerns, and the compounding ‘second shift’—while seemingly alleviated by the ability to ‘work from home’—are amplified by the sheer volume of work it takes to run a creative enterprise, a home, and a family.

From an operational point of view, AI can help female content creators scale their offering—editing videos faster, designing graphics, writing scripts, hooks, and social devices, analysing and predicting what’ll create their next algorithm-fuelling hook, and managing the admin of running an independent business. Solo creators have become full media studios overnight, so the possibility to scale without that much-needed access to capital becomes achievable.

But whilst AI solves for the struggle faced by female creators in their ability to scale and create equity, it’s also birthed a new challenge. Synthetic creators now exist in an industry highly optimized for algorithmic engagement. Manufacturing highly sexualized or idealized synthetic women feels like a predictable next step in an industry driven to monetize highly engaging visual content. Anyone can produce a stereotypically ‘perfect’ woman who is 100% synthetic (Cameron-James Wilson did this with Shudu Gram almost a decade ago). But what was once a rarity has become incredibly easy to scale and almost unrecognizable in its realism—well-dressed, perfectly curated, and monetized through brand deals executed without a human creator ever leaving the sofa. The way women are being represented online is rapidly slipping from human control. This creates complex challenges for women overall, but when you look at representation for women from global majority backgrounds, the negative impact is even more profound.

The algorithm fuels the creator economy. It’s always learning—training on what people are consuming, when they pause, for how long, how they react…a continuous loop serving users more of what they’ve paid attention to. Despite platform measures to block discovery by removing blatantly inappropriate content, if synthetic creators are only getting more convincing with subsequent evolutions of Gen AI and beauty is being homogenized by AI filters across platforms, how are we supposed to determine what real women look like anymore? The less control we have over what falls into our feeds, the more we expose a generation of young viewers to even more unrealistic and narrow standards for how women should look, think, feel, and behave.

I’m not going to go any further into the darker sides of deep fakes and digital exploitation here, as we could keep rabbit-holing for a long time on the impact of it all. Instead, I’d like to take pause and issue a call to arms. We’ve seen other Unmute contributors touch on the AI adoption and confidence gap.

Current trends show that women are roughly 25% less likely to experiment with and adopt gen AI tools than men.

If this gap persists, the development and deployment of synthetic media will lack balanced perspective, deeply impacting how women are perceived and how they perceive themselves in the future.

Used correctly, AI could be a massive accelerator for women in the creator space. As creatives and marketers at the forefront, we need to leverage it to actively dismantle pay bias, scale female-led businesses, and inject a fairer, more diverse female aesthetic into the branded content we produce. Most importantly, we’ve got a steep path to climb in re-training algorithms that are currently telling young girls they all need to look identical. For me, this mission is a personal one for my three little girls. They will grow up in a world where synthetic media is so normative they may never think to question its authenticity. It feels like a critical inflection point; how we consciously use human influence to train the synthetic output today will directly dictate how the next generation builds businesses and self-value tomorrow.

A few weeks ago we brought Dr. Laura Weis to speak to our LATAM leadership team. I expected to nod along to a few good slides and get some good conversation going but instead I left with the uncomfortable feeling you get when someone says the thing you've been circling for months but hadn't let yourself say out loud.

Her point, more or less: we have to pivot learning “on the job”. Not learning as a thing, you schedule. Learning as a thing that happens while the work happens.

I lead the center of excellence for talent & org effectiveness for the region for a company whose entire business is helping other companies transform. And here I was, sitting on a problem I couldn't solve with any of the tools I'd spent my career building.

The fight I kept losing

Here's the honest version of my role lately. We know we need to upskill our people, and fast, as AI is rewriting what good work looks like faster than we can build a course about it. By the time I've designed, reviewed, and delivered a polished module, the thing it teaches has already shifted.

For a long time I treated that as an engagement problem. “How do I make people want this enough to find the time?” What Laura's talk knocked loose was the realization that I had the problem backwards. It was never going to be an engagement problem. It was a design problem. As long as learning means stopping work to go do a separate thing, I am competing with the work and the work will win every time it matters.

The reframe

So I stopped trying to find people more time to learn and started trying to dissolve the line between working and learning altogether.

It sounds like a slogan. It's actually a fairly ruthless operational shift, and it changes almost everything about what my team does.

The clearest place it shows up is our AI rollout. We're putting these tools in people's hands anyway. For months I treated that as a “separate” track from upskilling. First we deploy the tool, then later we train people on the broader capability. But the rollout “is” the upskilling, if you design it that way. "Use this tool on your actual client deck this week, with someone on standby when you get stuck" teaches more than any module I could build, and it costs me nothing in anyone's calendar, because the deck was due anyway.

The tool becomes the curriculum. The real task becomes the classroom.

What this looks like when it's working

A few things start to change once you commit to the reframe.

You stop saying "go take the course" and start putting the answer where the problem is like a three-minute resource that shows up the moment someone hits the wall, inside the tools they already live in. People don't actually lack the time to learn. They lack the time to “leave their work” to learn. Close that gap and a lot of the resistance evaporates.

You also stop pretending the expertise lives in the center. In a year this fast-moving, my real experts aren't only on my team or on the Tech team, they're the early adopters scattered across the business who've already figured out how to make the new tools flow. My job is less to be the source of knowledge and more to build visible ways for those people to teach each other. A monthly "here's how I used AI to do this" demo. A channel where wins get shared. Office hours. It scales in a way that a central content factory never will.

And maybe the biggest shift: managers become the actual delivery system. The highest-leverage learning in any company already happens in a 1:1, in a stretch assignment handed over on purpose, in a project that gets debriefed instead of just closed out. My team's work is quietly moving from “making content” to “equipping managers to coach”.

The part I'm watchful about

I'd be lying if I said this was a clean win with no catch.

Flow-of-work learning only works if you protect a few real rituals around it: the visible peer teaching, one genuinely sacred block of time to experiment and apply, a manager who actually debriefs.

So I'm holding two things at once. Stop competing with the work and stop hiding behind the work. The discipline is in the difference.

Where I'm starting

Not everywhere. One capability, one team, proven first, then scaled as a pattern. Would love to retire my old favorite metric. Course completions tell me people “attended”. They never told me anyone could do anything differently on Monday. The new scoreboard must measure capability “applied” and the business outcomes that follow, or the gravity of the old way will pull me right back to building courses nobody has time to take.

I don't have this fully figured out. But for the first time in a while I feel like I'm solving the right problem instead of getting very good at the wrong one.

Thanks, Laura, for the nudge.

If you're wrestling with the same thing in your own team, I'd genuinely love to hear what's working. Reply or drop a comment.

For a long time, I did not think of myself as a real leader. Not in the way leadership should be.

In the beginning, I never came to the office thinking about governance, span of control, or reporting structures. I came in thinking about the problem. The client. The work. What we could build together that did not yet exist. I light up when I imagine something and feel a deep drive to make it real. Not just ideas. Ideas put to work.

That instinct has followed me across what others might politely describe as a “diverse” career: media innovation, creative agencies, client-side, tech. Rising into senior positions, my strengths started to feel like liabilities. Too broad. Too hands-on. Too passionate to be taken seriously as an executive. I began to wonder whether curiosity across disciplines was a weakness. The doubt caused real anxiety. Imposter syndrome. The impulse to try to fit in.

But when you spend 80% of your energy trying to be someone else, it is hard to be your best.

The classic playbook never seemed to fit

I have been through more leadership programs than I can count. I have held proper leadership titles: VP of Brand and Marketing Strategy at a major financial institution, associate professor at the Faculty of Arts, Head of Industry Global Brands at Meta. I learned influence leadership, governance, accountability.

And yet. Every time I sat in a room being taught what leaders should look like, I felt a gap between the archetype and myself. The leaders I longed to be like, operated with a kind of distance from the work. Composure over passion. Always asking the right questions but, when things got genuinely difficult, reluctant to give the answers.

I was the opposite. I got pulled into the work. Constantly building new models, testing prototypes, trying out new narratives. Not because I lacked structure, but because the detail was where the insight lived. Where the creative tension was. Where you could see the blueprint for what was worth building.

I compensated and overcorrected. Told myself to be more contained, more C-suite-like. Drop the sneakers, show up as expected. I treated my enthusiasm as something to manage.

“Nothing is as contagious as enthusiasm”

When I left Dentsu in 2010, my former CEO Lars Bo Jeppesen gave me a framed quote: “Nothing is as contagious as enthusiasm.”

It took me fifteen years to fully understand why he chose that gift. What I was good at, all along, was not the formal parts of management. It was leadership. Setting direction through the noise. Not by issuing mandates, but by innovation, imagining better work and articulating it with enough clarity and passion that people wanted to follow. Making others see a future they needed to be part of. Not because I asked them to. Because I could make them imagine it too.

That is a form of leadership. One that does not show up in most of the management frameworks.

M-shaped: A new shape to fit into

Dr Laura Weis recently wrote about “The M-shaped Orchestrator”. When I read it, I felt a form of belonging.

She argues that the T-shaped professional is no longer sufficient for the AI era. What we need are M-shaped professionals. People with multiple areas of deep expertise plus the connective ability to bridge disciplines. Integrators who span processes, people and thinking in ways that single-domain roles cannot.

This matters more now than ever. AI is increasingly capable of handling execution within a single domain. It can analyse, generate and optimise faster than any individual specialist. What it cannot do is connect unexpected dots between disciplines, navigate the human complexity of an organisation in motion, or imagine what is worth building that does not yet exist.

AI is raising the floor across the entire industry. Everyone can produce more, faster. The differentiator is no longer execution. It is the ability to see new paths and imagine what is worth building in the first place.

I had spent years apologizing for being too much of several things. But breadth, when it comes with an extended ability to see patterns across disciplines, is perhaps the essence of Laura’s Integration Axis: Not the specializations themselves, but the core ability to connect them.

We are not leading through change. We are leading in it.

A few weeks ago, I spent two days with Katerina Zacharia and Melissa Shilcock, testing and prototyping perspectives on what future-ready leadership actually looks like with our Nordic executive leadership team.

Katerina opened with something that stayed with me. We talk constantly about leading through change. But that framing contains a hidden assumption: that there is an end state. That if we are resilient enough and push hard enough, we will arrive somewhere stable. Somewhere done.

The future will not give us the comfort of done.

Technology is restructuring the marketing industry in real time, rewriting what skills are scarce and what roles are relevant. We are not navigating a transition between two stable states. We are operating in permanent change where the rules are being rewritten as we apply them.

“That is not leading through change. That is leading in change. And it requires a different skill set. The Strategic Orchestrator.” Katerina Zacharia

The leaders who will thrive are not those who can endure the struggle and reach the other side. Future-ready leaders are not defined by having the right answers or the most expensive suits. They are defined by the capacity to handle complexity with clarity. To learn faster than the environment changes. To create psychological safety and direction simultaneously, so that others can see a path forward.

When the herd has no map

When change is large, human instinct is to seek safety in the collective. The herd pools intelligence, distributes risk and, for most kinds of uncertainty, finds its way.

But what happens when the change is so fundamental that the herd itself has no map? When collective intelligence cannot find a shared direction, and the group just moves: Together but aimless. Anxious. Covering ground without going anywhere meaningful.

In those moments, someone must step forward. Not with a perfect plan, not with proven evidence, but with enough pattern recognition, enough cross-domain experience and enough accumulated instinct to say out loud: There is no map. But I have a very good bet on where we should go.

That capacity only becomes possible when you have spent years connecting dots across disciplines, staying curious enough to see what others miss, and staying close enough to the work to trust your own instincts when the data has not caught up yet. The capacity to commit to a direction before certainty arrives and articulate it with enough conviction that others will follow.

This is, I would argue, the most undervalued leadership skill of this moment. Not the person who waits for consensus before speaking. The one with the clarity to imagine a way forward and the experience as a builder to know how to get there.

The perfect fit for the misfit

After twenty years trying to fit in, fighting anxiety in boardrooms, overthinking my lack of executive presence, I finally feel I belong in leadership. In imagining the future of the industry. And, perhaps most importantly, belonging in myself.

My strength does not lie in looking like the assertive, contained, traditionally stereotyped male executive. It comes from persuasion, from driving engagement and trust, from strategic mindset and genuine curiosity. My development gaps are real. I am not naturally wired for structured planning and classic management. But for the first time, I can see the whole picture more clearly, and it is not incomplete. It is just a different shape than the shape I tried to fit into. In a world where AI reproduce the past, and by that, the leadership roles, as I thought they should be. We need to raise the agenda of what leadership can be. What it perhaps should be in the future. Fuel that future and its systems with new voices.

So, if you have ever felt too curious for your own good. Too broad. Too passionate about the work to stay comfortably above it. Too enthusiastic to be taken seriously as a leader.

I see you. And I think we may be exactly what is now being asked for.

Consider the training catalogue. Forty-seven modules. A completion dashboard. A certificate downloaded with mild pride and filed somewhere no one opens again. Buried in all of it is a single assumption: that knowledge, once deposited into a person, stays there and compounds like interest. That the organisation’s job is to fill people up and wait for results to improve.

It doesn’t work. And in an era when AI is recomposing roles faster than any annual learning agenda can track, it works least of all. The half-life of a skill is shortening. The training catalogue is getting longer. And the completion dashboard is recording excellent numbers against objectives that are increasingly pointless.

The gap tells you everything. Only 4% of organisations excel at using learning data to inform business decisions.¹ Companies with comprehensive training programmes generate 218% higher income per employee than those without.² That is not a training gap. It is a design gap.

What if the organisation itself is the unit of learning - not the individual, not the team, but the whole system?

What is actually being tested right now is not whether your workforce completed the AI literacy module. It’s whether your organisation can adapt at the speed AI is moving.

The Misdiagnosis

Most businesses are treating a design problem as a skills problem. The reasoning is understandable: skills are measurable, procurable, reportable. Design implicates structures, decisions, and incentives that senior leaders often had a hand in creating.

The awareness gap isn’t new. What’s new is the cost of it. Deloitte’s 2024 Global Human Capital Trends report found that 88% of leaders recognise that rebuilding trust and transparency is critical to their organisation’s success. Only 13% say they are leading on it.³ As the report itself concludes: knowing is not the barrier. Getting stuck in the doing is. That is not just a will problem. It is a design and measurement problem.

Organisations that stall on AI adoption tend not to be stalling because their people lack capability. They stall because workflows were never designed to incorporate AI output. Because decision rights are unclear. Because incentive structures still reward individual heroics over coordinated improvement. Because middle management has received no clarity about what to do differently, and considerable implicit pressure not to find out.

You can run every employee through an AI tools workshop and change none of that. The skills gap is real, but it is the symptom. The design gap is the disease.

Amazon’s Working Backwards⁴ process illustrates the alternative. The real learning happens because of the process itself, the iterative drafting of a PR/FAQ, the leadership reviews, the repeated cycles of debate and refinement which requires better thinking at every step. The system demands it. The process is the learning mechanism.

That is the model most organisations are missing, and it explains why so many AI transformation programmes, despite genuine investment in training, produce so little change in outcomes. Training informs. Embedded process transforms.

What a Self-Evolving System Actually Looks Like

The work is the learning environment. The moment you separate ‘training time’ from ‘work time’, you have already conceded the game. Knowledge that doesn’t connect to practice within roughly 72 hours dissolves. The brain sees no reason to consolidate information it has no immediate use for.

Pixar understood this before most businesses were willing to admit it. Every day, animators share their incomplete work not the polished version, but what co-founder Ed Catmull called the ‘ugly baby’ in open reviews with the full production team. The dailies are not a training programme. They are a feedback loop designed to surface what isn’t working while there is still time to fix it. Crucially, the group that provides feedback, Pixar’s Braintrust, a body of peer directors and senior storytellers has no authority to mandate changes.⁵ It offers perspective; the director decides. Feedback is deliberately separated from power, which is why people use it honestly rather than defensively.

The result: nearly $19 billion at the worldwide box office,⁶ and a run of creative and commercial consistency, particularly from 1995 to 2010, when 10 of 11 films scored above 85% critically⁷ that no other Western animation studio has replicated. Pixar built a system in which better work was the structural outcome.

Not learning alongside the work - learning as the work.

Culture is load-bearing infrastructure, not a values poster. You can design the most sophisticated learning architecture in the world and watch it produce nothing if the cultural conditions are wrong. Not because people are resistant - the usual diagnosis - but because systems have properties that operate regardless of what anyone intends.

An organisation where experimentation is quietly punished will not experiment, regardless of what the all-hands says about innovation. One where decisions are made by the person with the most authority rather than the best information will not learn from its data. One where mistakes are treated as performance issues rather than system signals will learn to hide mistakes rather than learn from them.

The diagnostic question is not ‘are our people willing to learn?’ Almost everyone is willing to learn when the conditions make it safe and worthwhile. The real question is: where does information get stuck? Where are feedback loops broken? Where does fear operate as a faster brake than any skills gap could be?

The diagnostic question isn’t whether your people are willing to learn. It’s where the system is stopping them.

The Leader’s Job

The version of AI transformation leadership that involves strategy offsites and well-produced internal communications about the exciting journey ahead is popular. It is also where most transformations go to produce excellent-looking slide decks and mediocre-looking outcomes.

The actual job is less cinematic. It is mostly about conditions. Who has the decision rights to change how work is done, and are they close enough to the work to know what needs changing? When something fails, does the system treat it as a signal or a verdict? When AI creates capacity, does that recovered time get reinvested deliberately, or does it dissolve into ambient busyness?

These are not rhetorical questions. They are design variables. And the leader’s job is to set them, not once in a strategy document, but continuously, through the decisions that tell people what the system actually rewards.

Satya Nadella’s transformation of Microsoft is the example that tends to end this argument. When he took over in 2014, he didn’t commission a new training catalogue. He changed the design variables. Stack ranking, which had made knowledge-hoarding a rational career strategy, was dismantled. The cultural anchor shifted explicitly from ‘know-it-all to learn-it-all.’⁸ Failure became information. Sharing became safe. The system, redesigned, began to teach itself. Microsoft’s market value has since grown from roughly $300 billion to over $3 trillion.⁹ The cultural architecture preceded the commercial performance

That is the pattern. When the conditions are right, learning compounds. The organisation gets faster, not just more informed. Each cycle of detect-signal-change-improve runs a little more efficiently than the last. The gap between organisations that have built this and those that haven’t doesn’t stay fixed. It widens. And it widens faster the more AI accelerates the underlying pace of change.

A leader who treats this as an HR workstream will resource it as a cost. A leader who recognises it as system design will treat it as the engine. That distinction, more than any training investment, is what separates the organisations that will look back on this period as the one where they pulled ahead.

The Measurement That Matters

If learning is the adaptive engine of the organisation, the question of how you measure it changes. Completions are out. What you are measuring instead is the organisation’s rate of update: how quickly it detects that something is not working, how efficiently that signal travels to the people who can change it, how reliably the change produces improvement rather than new problems.

AT&T’s $1 billion Workforce initiative transformed reskilling from a learning activity into a business lever. As documented in a Harvard Business School case study,¹⁰ the programme didn’t just track completions; it drove a 32% reduction in product development cycles and delivered a significant return on investment often cited in industry literature as high as $3 for every $1 spent by avoiding the substantial costs of external hiring and turnover.¹¹ They weren’t measuring learning metrics.

The moment an organisation stops asking ‘did they finish the course?’ and starts asking ‘what is different because of it?’, measurement becomes useful. And useful measurement, unlike completion data, tends to attract the kind of senior attention that turns learning from a line item into a strategic lever.

The organisations that invest in training and call it transformation will find, in a few years, that they have a well-certified workforce and an operating model that AI has quietly made obsolete around it.

The ones that have invested in building genuinely adaptive systems are already pulling away. They built systems in which better decisions are the structural outcome and that advantage accelerates with every cycle. The gap widens.

This is not a CPO issue. The operating model is a CEO decision. The ability to deliver on brand promise at speed when customer expectations are being reset by AI faster than most teams can brief an agency is a CMO decision. The return on system investment versus the mounting cost of capability obsolescence is a CFO decision. If this sits only in HR, it will be resourced as a cost and governed as a programme.

The question was never whether your people can learn. It was whether your organisation can.

Sources:,

¹ ATD (2025), The Future of Evaluating Learning and Measuring Impact: Improving Skills and Addressing Challenges

 ² ATD research, as reported by Forbes

³ Deloitte (2024), Global Human Capital Trends: Thriving Beyond Boundaries. Trust and transparency trend chapter

 ⁴Bryar, C. and Carr, B. (2021), Working Backwards: Insights, Stories, and Secrets from Inside Amazon. New York: St. Martin’s Press

⁵ Catmull, E. and Wallace, A. (2014), Creativity, Inc.: Overcoming the Unseen Forces That Stand in the Way of True Inspiration. New York: Random House

⁶ Pixar cumulative worldwide box office as of April 2026, derived from Box Office Mojo via Wikipedia

⁷ Rotten Tomatoes critic scores for all Pixar feature films 1995–2010, accessed April 2026. The sole exception is Cars (2006, 74%)

⁸ Nadella, S. (2017), Hit Refresh: The Quest to Rediscover Microsoft’s Soul and Imagine a Better Future for Everyone. New York: HarperBusiness. Stack ranking was formally abolished in November 2013 by HR chief Lisa Brummel, shortly before Nadella’s appointment

⁹ Microsoft market capitalisation at date of Nadella’s appointment, February 2014: approximately $303 billion (Bloomberg). Figure as of April 2026: approximately $3.1 trillion

¹⁰Kerr, W. R., & Fuller, J. B. (2019). AT&T, Retraining, and the Workforce of Tomorrow. Harvard Business School Case 820-017 ¹¹ Donovan, J., & Benko, C. (2016). Inside AT&T’s Radical Talent Overhaul. Harvard Business Review

A word spelled wrong, a punctuation mistake, a sentence with words in the wrong order. What was once considered unprofessional and sloppy is now proof of the human intent and a sign of caring about your professional voice.

In a recent Unmute article about how AI influences language, Radina Shipkovenska argues that we’re all becoming experts in humanwashing. The most obvious act of it being the removal of the em dash (I really miss being able to use dashes). In her article, Radina questions the evolution of our language: “I wonder if the next phase of humanwashing is the deliberate typo or a misplaced comma left in on purpose.”

I would argue the next phase of unpolished writing is already here. But there is another element at play than simply dressing your text up with a hint of human. Humanwashing will disguise the AI voice. But the LinkedIn post still reads as all the other posts in your feed. Your eyes still scan through the article without really taking it in. The polishedness of the arguments make your brain glide over them without friction – and without evoking any feelings.

But sometimes it’s different. Sometimes you hear the voice of the person behind the words when you read them. Sometimes you recognize the jumps in logic as the way your own brain works. Sometimes there actually is a human intent behind it. And the odd choice of a construct or word is exactly what makes you notice it. Then it’s not human-washing, but human-proof.

Persian carpet weavers have a long-standing tradition for adding a single knot tied wrong in an otherwise perfect field of knots. The theological explanation is that only God makes perfect things. That’s why it is also referred to as a humility knot. But the deliberate flaw has another purpose as well. It’s a sign of authorship. That this carpet came from a human hand, with a human decision embedded in it.

There are lots of examples and traditions through history of craftsmen and artists leaving this ‘maker’s mark’. From weavers and bookbinders including a hidden signature to developers leaving comments in code that no end-user will ever see. It is referred to as a maker’s mark because it draws the attention to the maker and their intention with what is made.

In professional writing, when human error was the only kind of error, a flaw in language was not a signature, it was just a flaw. But the moment AI started generating perfect texts, without bad days, or a subjective point of view, then the idiosyncrasies and mistakes of human writing began to carry a new meaning. And seeing these maker’s marks appear in professional writing implies a shift in what we value.

The most important shift this entails is the shift in valuing process over product. We are beginning to revalue interiority over output. We don’t care about whether the writing is clean, but whether it is inhabited. Whether there is a human sensibility behind it that made choices, including the choice to leave a rough edge, a flaw, or an unexpected use of punctuation? The polished surface, the hyper-professional voice, paradoxically, becomes the suspicious one.

Using deliberate flaws in writing as a stylistic choice is not new. One of the most famous examples is James Joyce’s Ulysses. The lack of punctuation was not automatic writing. It was a considered formal decision to make it sound like stream-of-consciousness. Deliberately mimicking the messiness of the human mind.

But not only is this new in a professional context, where considered and flawless used to be the ideal. It also serves another purpose. We’re moving from valuing the ability to make no mistakes to valuing the mistake in itself as a maker’s mark. The sleekness of AI craft makes us long for the messy and the unruly.

When the music video to Gener8ion and Yung Lean’s single Storm launched in April, it got millions of views in days. Not because it was the most seamlessly crafted music video, but because it was raw and messy, it was perfectly imperfect, and it showed the human emotion behind it. The dancers slightly out of synch, the geometric shapes they formed not quite symmetrical. For 7,5 minutes you couldn’t look away.

As AI collapses the distance between idea and output, the act of labouring through something, of leaving traces of effort and intent, becomes its own form of value. My prediction is we will see more of these deliberate flaws as proof of human intent. Not only in professional contexts, but across all types of creative output.

So, how do we tell the difference between real and performed humanness? Between humanwashing and human proof? The honest answer is that we might not be able to make that distinction. But the shift is widening the permission space for creators and makers to show themselves and their process in the product.

This article is written by me, researched and edited with Claude.
Photo by Cash Macanaya on Unsplash

Alt text: “A museum corridor with Braille text embedded into warm limestone walls in the foreground. In the background, a bold painting featuring multiple dark faces in rich reds, yellows and greens hangs on the curved wall. A digital screen is visible to the right. The space suggests a world where physical and digital languages coexist.”

Caption: “A space where touch, sight and technology speak the same language.”

I started out as an architect. At a crossroads, an opportunity came up to transition into creative technology. I took it, and it changed my life. But not the questions I was asking. Those just got bigger.

Architecture teaches you to read space. Who it was designed for. What it assumes about the person walking through the door. A corridor is never just a corridor. The height of a sign, the width of a doorway, the acoustics of a room — all decisions about whose experience matters, whether the designer thought about it or not.

I found those same decisions in creative technology, just faster and in ways much harder to see.

A few years ago, I ran a summer camp for New York City high school students exploring accessibility in a smart city — how technology and data shape the spaces we move through. My co-designer was visually impaired. She taught us to read with our fingers and navigate through space using our other senses — touch, sound, smell, spatial awareness.

Sitting there, feeling the dots of Braille under my fingertips, something clicked: this is another form of coding. A different way of translating the world. Not lesser, not a workaround, just a different language.

It opened a question I haven’t stopped asking since: if we can experience space through touch, sound and smell, how can we use technology to facilitate that? Not replace it, but open it up.

That question followed me back to London, and I started seeing the same problem everywhere. Not just in buildings, but in systems. The Underground, the signage, the crowds, the noise, the apps, the ticketing systems — every layer assumes the same person.

Walk into Tate Modern and the experience assumes you can read small print from a distance, that the crowds and noise are manageable, and that the layout makes instinctive sense. Everyone else is left to figure out how to navigate around it.

That’s not a London problem. That’s a design culture problem. And it’s being repeated in every digital and augmented environment being built right now.

The same industries that shaped the physical world are now shaping the digital one — bringing the same assumptions and gaps with them. Until the people most affected are part of the process, the same people will keep being left out.

All data reflects the world that generated it. Research consistently shows that diverse, representative input produces better outcomes for everyone. But that input is still too narrow, and those missing from it are the ones most affected by the consequences.

Neurodiversity is a powerful example.

Autism, ADHD, dyslexia and sensory processing differences are not problems for technology to solve. They are different ways of experiencing the world. When you design for someone who navigates without sight, finds a busy environment overwhelming, or needs to understand a space through sensation rather than signage, you ask different questions. And different questions lead to better answers.

The curb cut was designed for wheelchair users. Everyone with a pushchair, a suitcase or a bicycle uses it every day without thinking. Designing for the edges makes everything better.

And this gap is deeper for neurodivergent women — diagnosed later, represented less, and invisible in the data shaping these systems.

We’re already seeing what that absence looks like in the technology being built right now.

Generative AI doesn’t see women the way women see themselves. The image of us being built into these systems comes from data we didn’t generate, reflecting a version of us we don’t always recognise. Our spaces have been designed without us. Now our image is too.

As a Black woman, I’ve moved through spaces — professional and physical — that weren’t designed with me in mind. In architecture. In technology. In rooms where my presence wasn’t the assumption.

You develop a sharp awareness when that’s your experience.

You notice what’s been assumed.

You notice who’s been forgotten.

And what I keep noticing, in both the real and augmented worlds being built around us, is that the same people are being forgotten again.

Which is why what we pass on to the next generation matters.

I saw it at that summer camp. Young people who had never been asked to think about the human body, identity or who gets left out. Once they were, they couldn’t unsee it.

The next generation stepping into tech are building with AI tools from day one. But most are never asked to think about who those tools are built for.

That awareness has to start early, and it doesn’t come from a syllabus. It comes from who is in the room.

But it’s not just about the next generation.

Our biggest challenge isn’t only the biased systems we navigate. It’s the biased thinking we’ve taken on ourselves. We’ve been taught to follow the rulebook, even when it was written to exclude us. We’ve become so good at fitting into systems that weren’t built for us that sometimes we don’t even notice we’re doing it.

There is a toll to that. Of constantly being made to feel like you are the glitch, not the system being glitched.

That’s what it means to be muted.

Not always someone else silencing you.

Sometimes it’s years of quietly learning to do it yourself.

To every woman reading this: we know what this feels like. We have navigated industries, systems and rooms that didn’t account for us.

The most powerful thing we can do is trust our own experience. The part of us that knows when a space feels wrong, when a system isn’t fair, when a representation isn’t real.

That awareness is not a disadvantage. It is the qualification needed to design these environments differently.

What do we do on Monday morning?

In your next design review, your next sprint, your next team meeting, ask: Who are we assuming is the user? Who are we missing? Whose perspective haven’t we heard yet?

That’s the first step. That’s how you start to shift things.

My co-designer showed me that when you bring the right people in from the start, you don’t just fix a problem — you open up possibilities you wouldn’t have seen otherwise.

We are building the future right now.

We need to make sure we’re actually in the room — not just participating in the conversation, but helping shape it.

Adele Loye is a creative technologist at HEX Studio, WPP Production, exploring spatial design and inclusion across real and augmented environments.

I recently watched a TED Talk by etymologist Adam Aleksic about how LLMs are influencing the way we speak, particularly through the overuse of certain words shaped by training data.

It turns out “delve” is one of the more popular examples, largely because of how frequently people training and working with LLMs use it in their everyday language.

Once I started thinking about it, I began noticing it everywhere: in LinkedIn posts, presentations, even casual conversations. I used to use “delve” quite a lot myself and, admittedly, felt a little linguistically sophisticated every time I did. Now I’m starting to wonder whether those word choices were ever truly my own, or if I had simply absorbed them from the language around me.

Apparently delve is not the only word that has been overused, LLMs have a preference for words like – underscore, comprehend, bolster, the list goes on.

This recent finding got me thinking more broadly about language, and especially about speaking a second language as a non-native English speaker. I’m Bulgarian, and our educational system places a strong emphasis on foreign languages. I remember being taught that true language proficiency comes from understanding nuance - knowing the right word for the right context. And we take this seriously. I vividly remember using words like “cantankerous” to impress teachers during English exams or peppering an essay with “pernicious” for an extra flair.

These days are long gone. I hate to admit it but now I catch myself that I don’t think as long before I use a word – if it’s 80% there, it does the job. And it’s not because I don’t care, to the contrary, I care a lot but I’m rushing. I’m rushing to do things faster: to deliver work, to communicate more, to produce more, to be more.

And it’s happening with all of us, we’re all moving faster. AI has compressed the timelines around us, and somewhere in that compression, the space where careful language used to live has gotten smaller.

This isn’t an essay about AI ruining writing. I use AI regularly. As a non-native speaker, I’ve found it genuinely useful for moments when my thoughts outpace my English, when two (or three) languages tangle in my head and I need help finding the exit.

But I’ve been sitting with a question: have we lost something in the texture of everyday language? And if we have, do we even notice?

For the past two years, I’ve been working on research with Google exploring why people use AI: not the metrics of adoption, but the psychological, lived experience of it. One of the most consistent findings: AI raises the floor of human capability in domains where people aren’t experts, but it doesn’t lift the ceiling. AI might help make me an average writer but would never make me a Pulitzer writer and I think there is something beautiful about it. As humans we like imperfection, something that is not an algorithmic sequence of words that sound good together and is the proper way to write. I guess a prime example of this is the existence of haiku – it’s Japanese form built on three lines, 5-7-5 syllables, not designed to capture a story but a feeling. Not all needs to be perfect, it needs to be felt.

The same research surfaces another insight that I think is worth naming plainly: people are running the risk of becoming beige - of averaging out. If you spend any time on LinkedIn, you already know what this looks like in practice - a feed where every post sounds like it was written by the same person using the same formula. I initially thought it must be me, I must be reading a small portion of LinkedIn posts.

So I decided to test the hypothesis and asked an LLM to write a generic LinkedIn post about AI. It produced:

AI isn’t replacing people. People using AI are replacing people who don’t. The highest performers aren’t necessarily the smartest or most experienced anymore. The people who win in the next 5 years won’t be the ones with the most knowledge. It’ll be the ones who adapt the fastest.

You’ve seen this post or a variation of it. Roughly 40% of LinkedIn content is now AI-generated (the real number is probably higher) because what we’ve become quietly expert at is humanwashing: deleting the em dash, swapping out the words that feel too smooth, making the machine sound like us.

I wonder if the next phase of humanwashing is the deliberate typo or a misplaced comma left in on purpose.

The analogy that keeps coming to me is UGC (user-generated content) and the moment the aesthetic shifted from glossy photos to blurry, unfiltered images. The mess became the signal of authenticity. Maybe we’re heading somewhere similar with writing. Maybe the article with a punctuation mistake will start to feel more trustworthy than the article without one.

I don’t know what the future of writing and content creation holds, we are yet to live it but I’m an optimist by nature and I know that what human beings are surprisingly good at is correction. When the pendulum swings too far in one direction, we notice and eventually find our way back.

So here’s what I want to hold onto: the pressure to produce in milliseconds is real, but it isn’t fate. Thinking slowly is still allowed. Choosing the right word over the fast word is still an option. The pleasure of language - the particular satisfaction of finding exactly what you meant to say hasn’t gone anywhere. It’s just waiting.

And if you want to delve into a topic, by all means: delve.


* Header photo by Alexandra on Unsplash
* Graph was originally published on https://arxiv.org/pdf/2409.01754

A note before we start: There’s something a little ironic about using AI to write an article that’s partly about using AI less. I’ll own that upfront. I used AI to help shape and structure this piece, not to write it.

The thinking, the frustration, the arguments, the anecdotes: all mine. But honestly, AI has been one of the more meaningful things to happen to my career. It’s given me the confidence to put my ideas out there, including this one.

It has also taught me things. How to construct a sentence better. How to think differently about word choice. Writing clearly and at length no longer requires years of practice, and that matters.

The trick is knowing the difference between using AI to sharpen your thinking and using it to replace thinking altogether. This is my attempt at the former.

I really don’t believe everyone in the room understands what’s being said. And I’m increasingly worried that nobody is asking.

AI language is moving through companies at pace. Townhalls, pitches, briefings, workshops, company emails, presentations, thought leadership pieces. And the gap between what is being communicated and what is actually landing is getting wider. Not because people aren’t trying. But because the pace is relentless, the terminology is dense, and somewhere along the way it became easier to nod and move on than to stop and ask the question.

My role is in marketing & communications. I have to put in the work to understand this space properly. And even from that position I feel it. Can I really be the only person in the room who doesn’t immediately get it? The silence. The sitting on hands. The polite absorption of language that hasn’t yet been made real or relevant to the people receiving it.

I was talking about this recently, outside, drink in hand, with a friend, working through why I felt such an urge to write this article. She suggested I use the toddler agent. An AI tool that explains anything as if you are three years old. Brilliant, genuinely. But it also made my point exactly. Because a toddler agent is a private answer to what should be a public conversation. It means you never have to say “I don’t understand” to another human being. And the conversation you sidestep is the one where something new gets made.

That’s what worries me. And that’s why this exists.

A call to the industry - the art of the question is disappearing. And we need to bring it back.

Keeping up with AI, truly keeping up, not just scanning headlines but deeply understanding what it means and building real confidence around it, is one of the harder professional challenges right now. Not because the technology is impossible to grasp. But because it moves fast, the language moves with it, and by the time most people feel ready to engage with one idea, three more have landed to replace it.

And in that environment a habit is forming. A quiet, automatic habit. When we don’t understand something, we don’t ask a person. We ask the tech. When we need a term explained, we don’t turn to a colleague. We open a tool. We get an answer, alone, privately, without the friction and the unexpected places that a real conversation would have taken us.

We are building technology to replace the questions we should be asking each other. And I think that matters more than most people are saying out loud.

This is my call to be more inquisitive. Learn the definitions. Make them human. Carve out the time. Create the space. Put your hand up.

Because the present is hard enough to keep up with and the future is shifting faster still. The only way to navigate it, as teams, as organisations, as an industry, is together. Not each of us quietly consulting our own AI in our own corner, but collectively building understanding, sharing what we learn, testing what we think we know against the people around us.

For those of us in communications, this is the job inside the job. Keep pushing. Keep pulling the language back to something simple and human. Keep asking what it means for the actual person in front of you. Not just the business case. The human being trying to do their work well in a world that keeps shifting the ground beneath them.

The dictionary below is part of that work. A starting point, not a destination. Read it, use it, and then go and have the conversation it was trying to make possible.

The conversations are slowing. We need to make more space for them.

Everyone in this industry is navigating the same shift. The same new language, the same acceleration & onboarding of new tools, the same pressure to keep up while also getting the actual work done. Nobody has fully figured it out. The people who look like they do are just further along in asking the questions.

What this is about is keeping the conversation alive. Not letting it go private. Not letting the answer always come from a tool when it could come from the person sitting next to you, who might see it completely differently and change how you think about the whole thing.

So here is what I would encourage. After the next meeting, the next event, the next workshop: share what you took from it.

Not the deck. Not the link. What actually landed for you. What clicked. What still does not make sense. That kind of sharing, across teams, across clients, across disciplines, is how understanding actually builds. It is how best practice gets made. It is how the language gets human.

Build your community around this. Make the time. A standing ten minutes after the next big meeting where someone asks: what did you hear? What do you wish had been explained differently?

What are you going to do with this? Those conversations compound. They make the next one richer and the one after that richer still.

And while we are at it, I think about this beyond the office too. If this is what is happening in the professional world, if we are reaching for our phones instead of asking each other, what are we modelling at home?

My honest fear is that our children are watching us do exactly this. That they will grow up thinking the answer to not understanding something is to look it up alone, quietly, without ever having to say “I don’t know, can you help me?” to another person. We have to keep that door open. Ask the silly question at the dinner table. Let them see you not know something and figure it out together. Keep the art of conversation alive in every part of your life, not just the professional one.

Put your hand up. Ask the question. Keep asking, not because you don’t understand, but because the conversation is worth having.

To get the conversation started…

What is the last AI term you heard that you had to look up quietly afterwards? What do you wish someone had explained differently? Is there a word or phrase doing the rounds in your company that still does not quite make sense? Drop it in the comments and I will add it to the dictionary.

And if this resonated, share it with the person in your team who needs permission to ask the question. They are probably sitting on it right now.

Written by someone who is still working it out, same as you.

I’ve given and attended enough briefings to know how this goes. Someone puts up a slide. There’s a neural network graphic, maybe a gradient. The word ‘scalable’ appears at least 4 times. And somewhere in the deck, there’s a sentence that reads: ‘AI removes human bias from the equation.’

Everyone in the room nods silently. Either unaware or unwilling to admit that algorithmic bias powerfully reinforces and legitimizes existing inequities. It is as biased as you get.

Consider:

• +60% of websites are in English, only 20% globally, speak English fluently

• <1% of content is in African languages, yet Africans are ~17% of the global population

• One of the most-used AI training sources, Wikipedia, draws 80–90% of its content from Europe and North America

• Only ~15–20% of Wikipedia biographies are about women

If we accept AI outputs as fact, we risk allowing historical bias to define our future. This is critical moment where words and actions matter. Giving AI editorial control isn’t just lazy, it’s dangerous.

I get why it’s appealing. Clients want certainty. Everyone wants to limit liability. ‘The data decided’ is a clean exit from messy conversations. But data reflects the biases of those who create, fund, and distribute it, not objective reality. Algorithms amplify these historical inequities, turning them into system-wide decisions.

Accepting this false reality is not just pessimistic, it’s a brand risk, a cultural risk, and increasingly, a regulatory risk.

The plot thickens.

AI no longer trains on just human-generated data. It’s training on AI generated data. The outputs of today become the inputs of tomorrow. Which means bias doesn’t plateau. It compounds. What begins as skew becomes pattern. Pattern becomes norm. Norm gets laundered into objective insight. And once something achieves the status of objective, it becomes nearly impossible to challenge in a client meeting or a courtroom.

We stop asking if it’s right. We start defending why it is.

So, how is the industry responded to this threat?

‘We need more diverse data.’ Sure. But data collection is itself a power dynamic. You don’t solve representation by extracting more from communities that have historically been underserved. That’s not inclusion. That’s better-informed extraction with a DEI footnote.

The real question isn’t who’s in the dataset. It’s who decides what gets collected, how it’s interpreted, and what it ultimately gets used to justify. Until that changes, more data doesn’t fix the system. It just makes the bias more statistically confident. Which, if you think about it, is worse.

So, what does ‘doing it right’ look like?

A few thoughts worth pushing for inside your agency, and with clients:

Audit the Inputs, Not Just the Outputs.

Most agencies review outputs (creative, targeting, messaging). That’s too late. Responsible stewardship means interrogating:

• What datasets are being used

• What’s excluded (geographies, languages, income brackets, behaviors)

• What proxies are quietly standing in for sensitive attributes

Bias isn’t a glitch in the output. It’s a design choice upstream.

Build ‘Bias Stress Testing’ Into Go-To-Market.

Before anything launches, pressure-test it like you would a financial model.

Ask:

• Who does this system systematically underserve?

• Who does it over-prioritize?

• What happens at the edges (non-majority users, atypical behaviors)?

Make this as standard as QA or brand review. Not a workshop. A gate.

Redefine Performance Metrics.

AI systems optimize for:

• CTR

• Conversion

• Efficiency

All of which can reward bias. You, need to expand the definition of performance to include:

• Distribution fairness (who is reached vs excluded)

• Outcome parity (who benefits vs who is penalized)

• Long-term brand trust impact

If bias improves short-term performance, the system will choose it every time unless you redefine what ‘winning’ means.

Diversify Data Partnerships, Not Just Teams.

Agencies talk a lot about diverse talent. Less about diverse data sources.

Push clients and partners to:

• Incorporate non-Western datasets

• Include multilingual and non-dominant market inputs

• Balance commercial data with cultural/contextual data

If the data is narrow, the thinking will be too no matter who’s in the room.

Shift the Narrative: From Ethics to Advantage

Responsible AI sounds like a constraint. Reframe it as:

• Better data = better targeting

• Broader inclusion = expanded market opportunity

• Reduced bias = stronger long-term performance

This is not about being good. It’s about being right and winning because of it.

Build accountability into the system.

Transparency reports don’t shift incentives. Real consequences do. The EU AI Act is a start. Watch it carefully and get ahead of it, because your global clients already are.

Create a ‘Bias Escalation Path’ (Like Brand Safety)

Brand safety has escalation protocols. Bias should too.

• Who flags it?

• Who investigates it?

• Who has authority to pause spend, retrain models, or kill work?

If there’s no escalation path, bias becomes everyone’s problem, and therefore no one’s responsibility.

The real threat isn’t a rogue algorithm. It’s the perfectly functioning one.

The popular fear is AI gone rogue; autonomous, uncontrollable, dramatic. That makes a great doomsday movie but it’s not the actual risk profile we’re operating in. The real danger comes from the algorithm that operates precisely as intended, automating and entrenching bias while presenting it as fact.

No alerts. No visible failure mode. Just millions of compounding small decisions, all pointing in the same direction, quietly determining who gets access, who gets seen, who gets the offer, and who gets filtered out.

We still have a window. It’s narrowing.

The systems being built right now will define the decision-making infrastructure of the next decade. Once embedded at scale, it won’t just influence outcomes; it will determine which outcomes are even imaginable.

Our industry has always had a complicated relationship with the question of who we’re really building for. AI doesn’t resolve that question. It amplifies it, accelerates it, and eventually makes it very hard to walk back.

Start here: design systems that reflect the full complexity of human reality, not a flattened, convenient version of it. Because if we don’t do this deliberately, we are not building intelligence, we are industrializing bias. And worse, we are disguising it as objectivity.

We need to talk about the narrative that’s quietly shaping how the world thinks about AI — and why it’s wrong.

“AI will augment the human.” You’ve heard it. You’ve probably said it. It’s the sanitised, optimistic framing that makes the AI revolution feel safe and managed. But listen carefully to its grammar. AI is the subject. Humans are the object. We are — yet again — positioned as the passive beneficiary of a power that sits elsewhere.

For women, this should sound familiar. The passive beneficiary. On the receiving end of a system built and run by others. It’s a power dynamic women have been navigating for centuries. So here’s the provocation: what if, rather than accepting that framing, we flip it entirely?

It’s not AI augmenting the human. It’s the human augmenting AI. And the human best equipped to do that — is female.

This isn’t wishful thinking. It’s evidenced.

My children’s school recently reframed its approach to preparing students for the future of work. Amid a generation being told AI threatens their careers, the school’s opening message was quietly radical: the skills you will need most are creativity, intuition, judgement, and empathy. It stopped me in my tracks. Because if those are the skills that will define professional value in an AI-powered world, then women — who have long been told those same skills were too soft to count — are sitting on a gold mine.

The data agrees. The World Economic Forum’s 2025 Future of Jobs Report identifies empathy, active listening and creative thinking among the most critical human skills going forward — and finds they have zero substitution potential from current AI. No algorithm is coming for your emotional intelligence.

Academics Roland Rust and Ming-Hui Huang, in their landmark book The Feeling Economy, argue that as AI takes over thinking tasks, the competitive advantage shifts entirely to feeling: intuition, empathy, relationship intelligence. Their conclusion is striking — in the Feeling Economy, women are inherently advantaged. A 2025 peer-reviewed study in Frontiers in Psychology backs this up, finding that female intuition demonstrably leads to superior leadership performance in conditions of uncertainty and upheaval. It’s worth noting that much of what we call ‘female intuition’ is substantially shaped by socialisation — by decades of women being required to read rooms, manage relationships and navigate systems not built for them. Whether nature or nurture, the capability is real, it is developed, and in the AI era, it is valuable.

And it’s already showing up in leadership behaviour. A 2026 study by Chief and The Harris Poll — Beyond Speed: How Women Are Defining the Human-Agentic Workforce — surveyed over 1,000 senior women leaders and found that 68% use AI primarily to amplify human talent rather than replace it, and 85% believe organisations that invest in both AI and people will outperform those betting on technology alone. Women aren’t behind on AI. They’re ahead on the question that actually matters: what should humans do that AI cannot?

Here’s the uncomfortable tension, though.

The WEF’s own data shows that women’s representation in senior leadership fell for the third consecutive year in 2024 — from 38% to 36% of leadership hires. Women remain just 31.7% of senior leaders globally, despite making up 42% of the workforce. So even as the skills women have built become the most strategically valuable in the room, the doors to the rooms where those skills are needed most are — still, stubbornly — not opening fast enough. That gap between value and access is not a reason for pessimism. It’s a reason for urgency.

The skills that were historically dismissed as soft — labelled feminine, and therefore lesser — are the very skills that make humans irreplaceable in an AI world. Empathy cannot be automated. Intuition cannot be prompted. Judgement in conditions of ambiguity is not a feature you can download.

For too long, the ‘female way’ of leading was treated as a workaround: tolerated, occasionally celebrated in a diversity report, but never truly valued as strategic advantage. That era is over.

AI hasn’t just changed the future of work. It has, without meaning to, revalued an entire set of human capabilities — and women have been quietly building those capabilities their whole careers.

So stop waiting to be augmented.

Start augmenting.

Early on, I tried to build an insight agent. Gave it the problem I was trying to solve, connected it to data, told it to find insights, waited for something brilliant.

What I got back were gems such as “with almost 90% of UK adults now shopping online, brands must rethink their digital-first strategy to meet consumers where they are” or “authenticity has become the defining currency of brand trust for Gen Z.” Statements made with the kind of confidence that might trick you into thinking they aren’t utterly useless.

It took me a minute to realise the problem wasn’t the agent. It was me. I’d asked for insights without ever defining what an insight actually is. I hadn’t explained what differentiates a genuine insight from an observation, a trend from a truth, a data point from something you can actually act on. I’d outsourced the task, and in doing so, I’d outsourced the judgement. The agent defaulted to what the model training data – the internet – thinks an insight looks like, which is often just numbers with a narrative wrapper.

That experience changed how I think about agents entirely. It taught me that the question to start with isn’t ‘what task do I want my agent to do?’, it is ‘which decisions am I giving away?’.

The hidden decisions

The word “agent” comes from the Latin agere, meaning to act. But, agency isn’t just the ability to act. It’s the ability to decide and then act. When we give AI agency, we need to be aware that we are not just handing over tasks. We are handing over decisions. And, those decisions aren’t just the big, obvious, high-stakes ones. They’re small, buried deep in the design, and not always clear at the outset, such as how to interpret a specific word, apply an ambiguous instruction, or determine what “good” looks like when nobody has said otherwise. The inherent risk is because AI is extraordinarily good at sounding intelligent, those hidden decisions almost never get questioned.

My insight agent wasn’t broken. It was making a perfectly reasonable decision about what an insight is, based on what it’s training data has taught it. The problem is that the internet is wrong about many things worth being right about, and I hadn’t told it otherwise.

Once I started seeing agents as decision makers, it became impossible to unsee it. Let’s say a brief lands, asking us to develop a launch campaign for a brand entering a new market. The team maps the work out as steps in a process, and point agents at the tasks in each step. But, look a level deeper. A successfullaunch isn’t just about steps in a process, it comes from dozens of decisions, each shaping the next. Who do we target? What behaviour needs shifting?

Which insight do we build on? Which creative idea do we back? Does our execution meet guidelines? How do we allocate our media budget? Some of those decisions have known answers. Some have answers we can confidently predict based on data, experience and expertise. And some have no right answer at all, just better or worse options, and a human who has to live with the consequences of choosing.

That spectrum tells us something important. Not all decisions should be delegated equally. Where there’s a knowable answer, agents can be fast, consistent, tireless partners. Where expertise is needed, agents can help, provided they’ve been built with that expertise genuinely baked in. But where there’s no right answer, where it comes down to taste, conviction, the kind of judgement you build over years of doing the work, that’s not a gap in the technology. That’s the point of the human.

Agency without accountability

The reason decision-making rights matter should be an obvious one: decisions, no matter how small, have consequences. The inherent challenge here is that while we give agents agency, they do not have accountability. They can act on our behalf, increasingly autonomously, but the goals aren’t theirs. It is no skin off an agent’s back if the strategy is brilliant or mediocre. They don’t feel the weight of a recommendation that changes a client’s direction. They have no real stake in the outcome.

This is one of the reasons you’ll hear me using the phrase “humans at the helm” instead of “humans in the loop.” Being in the loop is passive, it’s the nicety you hear at the end of a meeting when someone promises to keep you loosely informed. In agentic, it means you’re checking the machine’s homework, approving outputs, and taking a backseat role in an automated process. Being at the helm is fundamentally different. It means humans set the intention: what problem we are solving, why it matters, and what success looks like. It means humans make the calls that don’t have right answers. And it means humans own what comes out the other side, regardless of how much of the work was done by agents.

That ownership can’t be abstract, it has to be both specific and explicit. Which decisions in this workflow are mine? Which are the agents? Where do we collaborate, and what happens when the agent hits something it wasn’t designed for? The best teams I’ve worked with over twenty years in strategy have always had this clarity, not written on a wall somewhere, but embedded in how they operate. Everyone knows who decides what, when to escalate, and where the buck stops. Human-AI teams need the same discipline. Arguably more, because when an agent gets it wrong, it does so confidently and at scale.There’s no furrowed brow to signal doubt, no hesitation to suggest uncertainty, just a polished output that looks so good we stop questioning it.

Knowing where to stand

We talk a lot about taste and judgement being the irreplaceable human contribution in an age of AI, and I think that’s right. As answers become more accessible, the value shifts to the questions we ask and the judgement we bring to the answers, especially when there’s no inherently right answer. There’s a harder skill underneath this, though. It’s not just having good judgement; it’s knowing in which decisions your judgement needs to be applied.

Because you can’t apply it everywhere. There are too many decisions in any piece of work, and some of them genuinely are better handled by agents. The real skill is in reading the chain of decisions and knowing which ones are yours, which ones are theirs, and which ones you need to make together.

That’s what I got wrong with my insight agent. I didn’t lack judgement about what a good insight looks like, I just didn’t realise I needed to apply it before the agent ran, in the design itself, in the criteria, in the definition of what I was actually asking for. The decision I gave away wasn’t “find me an insight.” It was “decide what an insight is.” And that one was mine to make.

Last week, I had an amazing opportunity to showcase some of the Physical AI innovation works WPP have been doing at Google Next. It got featured in 2 key moments, a Spotlight session - What’s next in AI infrastructure: Scaling for the agentic and physical AI era, by Mark Lohmeyer, VP & GM, AI and Computing Infrastructure at Google Cloud and Stephan Pretorius, CTO at WPP - and the Google Cloud physical ai booth at the Expo floor.

Preparation: planning for the unexpected

Before anything reaches a stage or an expo floor, there is a long stretch of preparation that doesn’t look dramatic from the outside but defines everything that follows. Much of Q1 was spent building the reinforcement learning pipeline, node workflows, and rendering key moments for the film.

My role spanned across programming, mechanical engineering, and network engineering for Spot - the quadruped robot by Boston Dynamics fitted with a Sony camera with Cooke lens, built for professional filming. The goal wasn’t just to make the system work, but to make it reliable in real environments, where timing, connectivity, and physical behaviour all need to align.

Events like Google Cloud Next add another layer of complexity: dense crowds, heavy network interference, and long hours of repeated operation. A lot of the preparation went into identifying potential weak points and thinking through how the system would behave when conditions weren’t ideal.

Spotlight session: Spot on stage

During the Spotlight session, the focus was on new infrastructure announcements - including TPU 8t and TPU 8i - and how this next wave of compute is shaping not only large-scale AI systems, but also the future of Physical AI.

As part of that story, Spot joined the speakers on stage. From the audience, it looked like a natural extension of the ideas being presented. From backstage, my focus was on making sure the system behaved exactly as expected so that the moment on stage could feel effortless.

Expo floor: in the middle of conversations

On the Expo floor, Physical AI felt less like a vision of the future and more like a tangible reality. People didn’t just walk past. They stopped, watched, filmed, and asked questions, connecting the the new cloud capabilities unveiled at Next to tangible Physical AI solutions.

As part of the Google Cloud G4 demonstrations for training Physical AI systems, Spot was showcased alongside a G4 server powered by the NVIDIA RTX PRO 6000 Blackwell Server Edition.

Being there, answering questions while also operating and adjusting the system, made Physical AI feel very immediate & tangible — not as an idea being presented, but as something people could see, ask about, and relate to directly.

An unexpected moment

One moment that stayed with me happened on the expo floor while I was actively working on the system with my tools, rebuilding the camera system on Spot.

Someone stopped, watched for a moment, and said:
“I had to say. I loved that you are the one behind this.”

I was in a room filled with decision makers moving through the expo, and in that context, being a young female engineer working directly on the system gave a different perspective on the experience, not as an observer, but as someone maintaining and operating the technology while conversations around it were happening.

What stayed with me wasn’t the comment itself, but the setting. These kinds of events bring together people who build, decide, and influence where technology is going, and being part of that environment while doing hands-on engineering made the connection between ideas and reality much more immediate.

It also reinforced the value of more engineers in hands-on roles — including young engineers and more diverse voices — having access to these spaces, not just to see what is being announced, but to understand how those systems come together in practice.

The promise of artificial intelligence is vast, but it’s true power lies not merely in its algorithms, but in the collective human intelligence that shapes it. For 15 years, in my professional capacity leading large-scale transformation programs, I’ve observed a foundational truth: the most robust, innovative, and sustainable outcomes consistently emerge from a fusion of diverse voices, lived experiences, and distinct points of view. This isn’t just a professional conviction - it’s a lived one. Growing up as an engineer in a family of artists - musicians, painters, designers - I was always the oddball; the logical one in a sea of creatives. That tension became my greatest asset: it taught me to listen without judgment, to test every perspective with rigor, and to understand that what I couldn’t imagine, someone beside me invariably could.

I am convinced this will be even more important in the future.

Watching my 15-year-old son intuitively leverage AI in his daily life only deepens this conviction. His fluency with these tools doesn’t diminish his capacity for human connection; it sharpens it. The future demands a collaborative dance between human ingenuity and artificial capability - a vibrant dialogue between generations and disciplines. The power of passing on human experience, wisdom, and nuanced understanding still profoundly outweighs pure algorithmic speed.

Yet a seismic shift is underway, challenging our very understanding of team dynamics and organisational structure. Data points that landed this past month signal a critical threshold: the idea of the “one-person unicorn” has become tangible reality. Matthew Gallagher built Medvi, a telehealth company, using AI to write code, generate ads, and run operations with a single employee, achieving year-one revenue of $401 million, now tracking $1.8 billion.

We are not merely heading toward a future of work; we are inside its first round right now. Jack Dorsey’s restructuring of Block echoes this revolution. His argument is structural, not financial: the traditional corporate hierarchy was a structure invented to route information through organisations too large for any one person to oversee. That problem, he argues, is now solved. Block will operate on three roles only: Individual Contributors, Directly Responsible Individuals, and Player-Coaches. Everything the hierarchy used to coordinate, the system now handles.

Scale Without Lost Purpose

This raises a critical question: In a world where a single individual, augmented by AI, can achieve unprecedented scale, is collaboration still a benefit - or even a skill?

My answer is unequivocal: yes. But its nature transforms profoundly.

My experience reveals a ratio I return to repeatedly: 20% of AI’s value comes from the technology itself; the profound 80% is unlocked through the radical redesign of how work actually flows. Agentic AI doesn’t simply make existing processes faster - it renders many obsolete and creates entirely new ones. This isn’t about automating the past; it’s about inventing the future.

In this landscape, winning isn’t about having “better AI.” It’s about asking better questions - specifically, the questions nobody else is asking. A single individual might execute at scale, but who guides the purpose, defines the ethics, and asks the truly differentiating questions that ensure this scale serves humanity’s best interests? This is where diverse human intuition, experience, and the courage to defy norms becomes not just valuable, but irreplaceable.

AI can and should liberate us from the predictable and the procedural - freeing human collaboration to focus precisely where it matters most: defining purpose, building trust, and transferring the kind of wisdom no model can yet replicate.

The question isn’t whether AI will surprise us with unexpected capabilities. It already has, and will again next week. The question is whether the diversity of the human minds stress-testing them is keeping pace.

The Mindset That Meets This Moment

This is precisely where diverse human perspectives become non-negotiable. My own approach to co-working with AI agents rests on principles I apply as rigorously to myself as to any challenge:

• Question every norm. The most powerful prompt you can give an AI agent is the assumption you’re most frustrated by in your own field. Constraints aren’t limitations - they’re creative direction.

• Think without limits. Design from radical abundance first. Remove budget, technology, and feasibility constraints entirely, then work backwards.

• Iterate over perfect. Don’t design the final thing. Design the first thing - good enough to argue about, or for. And then continuously improve.

• Embrace open-mindedness. The smartest move is often birthed from the dumbest question (or the question most are afraid to ask). Be as curious as a pre-schooler, and as deliberate as an engineer.

These aren’t productivity tactics. They are the mindsets that ensure powerful AI agents are directed toward ethical, meaningful, and genuinely valuable outcomes - outcomes that reflect the full complexity of the world they’re designed to serve.

The future of AI isn’t defined by its algorithms. It’s defined by the intuition, empathy, and discernment of the humans who guide it - especially those who understand that the real transformation isn’t technology replacing humanity. It’s technology enabling more of us to be more experimental, more imaginative, and more uniquely, irreplaceably human.

And because AI shaped without diverse voices doesn’t just underserve the world - it quietly encodes the very limitations we should be dismantling.

I enjoy reinforcement learning – it appeals to my analytical nature. At its core, it’s a rather simple idea: you can teach a computer something by letting it try a task many, many times. If it does something right, offer a reward; if it does something wrong, give out a penalty. As the machine continues to try the task over and over, it reflects on its rewards and penalties. As it makes choices, the options become weighted based on its past experiences. Eventually, it figures out the right choices for a series of decisions to perform the desired task.

Currently, we are teaching a quadruped robotic dog how to walk over various surfaces with the added challenge of a custom camera rig mounted on its back and arm. If we create the right environment and run the right number of tests, this seemingly impossible task becomes possible. As an engineer, I find peace in the ability to take a problem and script an algorithm to create a solution. There’s joy in knowing problems have logic-based solutions.

Then my baby became a toddler….

For the uninitiated, when you have a baby, the first question after “boy or girl?” is “how’s the sleeping going?”. For the latter, I was always able to proudly (smugly?) answer that we had a wonderful sleeper – and we did! Clearly, some higher power decided that my husband and I were out of our depths and that we should at least have a baby that sleeps, but I would also like to think that our research and logic-based approach to a schedule helped. But after your baby turns 1, the most common question becomes, “Are they walking yet?”. My previous sense of pride deflated as I had to repeatedly admit, “Not yet.” As the months after my son’s first birthday ticked by, I started to wonder if his delay reflected on me. He was perfectly capable of walking but, if he wasn’t holding my hand, he would crumble to the ground – not unlike our robot dogs when they stumbled in the simulation.

And the robot dogs weren’t faring much better. Each test takes approximately 15 hours to complete. But after each failure, my colleague Jakub Molis and I could analyze the code, isolate variables we wanted to investigate, and run controlled experiments across our machines. It required patience, but if there’s anything that parenting a toddler can teach you, it’s patience.

However, patience and analytics seemed to have met their match with my son. At nursery, the staff said that we needed to build up his confidence. I had no idea what to do with that piece of advice, but I started to question the amount of time he spent at daycare. Was his lack of confidence due to me going back to work? Had I already squashed his chances of having high self-esteem because of my own ambitions to continue my career during motherhood? I think mothers are more capable of understanding reinforcement learning because we’re used to running hundreds of thousands of simulations… and that’s before breakfast. But, unlike our reinforcement learning algorithm, I found myself assigning more penalties than rewards.

My analytical mind went to work – could I calculate my time with him and see if there are improvements, awarding myself for his successes and accumulating penalties for his supposed ‘failures’? Would I learn the right choices and eventually ‘have it all’?

Unfortunately, an algorithm has no way to codify a Tuesday where I decided going straight to work and coming home in time for dinner would be the optimal way to split my time. That plan was derailed because that morning, my son did not want to wear trousers and the only way to calm him down was a full singalong accompanied by an extended round of cuddles. There is no calculation for a little human learning big feelings.

Meanwhile, reinforcement learning was working wonders with the robots. We made progress with each run of the algorithm. Eventually, Jakub sent me a video of the simulation running with each robot dog expertly navigating various terrains without a single stumble. They certainly didn’t need to hold my hand! Professionally, it was a wonderful achievement, but personally I wondered if the dogs walking before my child was somehow symbolic of me choosing a career over being a more present mother.

I think the real journey in crafting an algorithm to “have it all” is the slow realization that such an algorithm doesn’t exist. Some days one thing will get priority over another, and most days will be a bit messy – emotionally or quite literally. Today I made it out the door without any tantrums (there were actually two, but my husband handled both), but there is definitely a small but noticeable milk stain on my trousers. Jakub had made an interesting discovery when the robot dogs refused to take any steps forward. We needed larger rewards for the robots moving relative to the reward for standing still perfectly. Valuing perfection over movement resulted in everything remaining stagnant.

And my son finally started walking. My best friend was telling me a story when we both realized he was standing all by himself and then he simply walked over to me. He now walks with joy (and confidence!), revealing all the ways my husband and I failed to babyproof our flat. Maybe this is my new algorithm: rewarding me with smiles and giving penalties by finding every cabinet that isn’t child locked.

AI is no different. Except this time, what’s being rearranged isn’t an industry or a format. It’s something closer. It’s the story we’ve been telling about ourselves — that intelligence is what makes us matter. That the thinking is the thing. We built something that thinks faster, wider, and without needing sleep. And instead of that being the end of the story, it turns out to be the beginning of a more interesting one.

This article is about what comes after we let go of the story of loss. Not what AI takes from us. What it moves. And why, somewhere in that rearrangement, being human turns out to matter more than we thought.

Where We Still Matter

I saw something recently that didn’t look like much.

A person walking through a grocery store with a camera strapped to their chest. No commentary, no edits. They opened a fridge, picked something up, stood there for a second, then moved on. It felt almost pointless to watch. Which is how I realized it wasn’t meant for me.

The footage was being collected to train an AI system. Not to recognize objects or predict behaviour. To understand what the world looks like from where a person stands.

That detail stayed with me. Not the technology. The fact that it was necessary.

Here’s what’s easy to miss in all the noise about intelligence. The systems we’re building are extraordinarily capable — they can search, plan, write, analyse, and execute across digital systems with a kind of autonomy that would have seemed implausible five years ago. But they keep running into the same wall.

They cannot smell a room. They cannot notice the small tension between two people in a meeting. They cannot stand in an aisle and register what’s actually there: the gap on the shelf, the handwritten sign, the thing that doesn’t match the database. They can work with representations of the world. They cannot be in it.

So, when an agent reaches that boundary, it does something quietly remarkable. It reaches for a human. Not as a colleague. As a witness.

This is the shift worth paying attention to. It is not the automation. It is the rearrangement.

For a decade, the AI conversation has focused on capability. On how powerful the models become, how many tasks they touch, which jobs sit closest to the edge. Those are real questions. But they might not be the most interesting ones anymore.

What’s emerging is something more structural. AI may not simply replace human work. It may redefine where humans sit. Who designs the systems, who operates inside them, and who ends up as the sensing layer that machine intelligence cannot replicate. The role of the humans will not be replaced. But rearranged.

And here is where it gets interesting. We built something extraordinarily intelligent. And it turns out that intelligence, on its own, cannot be anywhere. It cannot arrive. It cannot witness. For that, it needs us.

Which makes you wonder whether intelligence was ever really the thing that made us matter.

We assumed it was. The whole anxiety about AI rests on that assumption: if something thinks better than us, we lose. But what if we were measuring the wrong thing all along? What if the thinking was never our most valuable contribution and it took building something that only thinks to finally show us that?

That’s not a loss. That’s a clue.

The First AI-Native Generation

It happened again this morning.

An idea arrived, fully formed, felt like mine. And then, a second later, the doubt. Did I think of that? Or did I think with something, and the line between the two has just quietly disappeared?

I’ve started noticing this more. Not with alarm. With curiosity. Because it’s not new, exactly. It just feels new because we don’t have language for it yet.

When Google became part of everyday life, something shifted in how we related to memory. You no longer needed to hold every fact in your head. You needed to know what to search for, how to evaluate what came back, when to trust it. Memory didn’t disappear. It extended. Outsourced some of its load, it kept what mattered most.

AI is doing something similar, but not to memory. To thought itself.

It’s not replacing the thinking. It’s sitting beside it. Suggesting, extending, challenging in real time. And the longer you work with it, the harder it becomes to draw a clean line between where you end, and it begins.

For most of us, that still feels strange. We notice it because we remember thinking without it.

For the generation growing up now, there may be no before.

Every generation absorbs its technologies so deeply they eventually stop feeling like technology. Millennials grew up discovering that the world’s information could be searched and that changed not just what they knew, but how they related to not knowing. Gen Z grew up where online and offline life collapsed into a single continuous experience. They didn’t adapt to that world. They just lived in it.

The next generation may grow up with something different again. Not a tool they reach for. A presence that’s simply there; while they’re thinking, learning, struggling with an idea at midnight. Always available, responsive.

Children growing up now will not experience this as remarkable. They will experience it as normal.

And that changes everything. Not in the dramatic way we keep imagining, but in the quiet, foundational way that only becomes visible a generation later.

The question I keep returning to isn’t whether this makes us smarter or lazier. It’s what it does to the texture of thought itself. What it means to arrive somewhere intellectually and not be entirely sure of the path you took to get there.

But here’s what I’ve come to think: that uncertainty isn’t necessarily a loss. Google didn’t make us worse thinkers. It made us different thinkers — more connected, faster to orient, less precious about holding everything ourselves. The friction it removed was mostly friction we didn’t need.

Maybe this is the same. Maybe the blur between your thinking and AI’s is less like a boundary being erased and more like a capacity being extended. A new kind of cognitive reach.

We just don’t have a word for it yet.

Ambidextrous describes equal facility with both hands. What do you call someone who thinks fluidly with both human and artificial intelligence? Ambimental. Cognitively bilingual. Dual-native. Synthient.

None of these are right yet. Which is probably the point! We’re naming something that doesn’t fully exist yet, in people who haven’t grown up yet, with implications we can’t fully see yet.

So, I’ll leave it open.

If you work in language, in culture, in the space where new things need new names, what would you call it? The first generation to think in two registers at once deserves better than a footnote.

They might be the most interesting humans we’ve ever produced. Or they might just be us, extended. Probably both.

The Last Skill Nobody Is Talking About

Something strange is happening to our attention.

Not in the way we usually talk about it: the scrolling, the dopamine, the shrinking focus spans. Something different. Something quieter.

We’re scrolling past good things.

Not bad things. Not lazy things. Things that are genuinely competent, considered, well-made. And we keep moving. Not because they failed. Because nothing about them made us stop. There was no edge, no commitment, no sense that someone had made a real decision about what this should be.

Everything is good now. That’s the problem.

AI didn’t cause this exactly. But it accelerated something that was already true: that the hardest part of making anything was never the making. It was the choosing.

Before AI, scarcity forced decisions. You couldn’t generate ten directions by Tuesday, so you committed to one. You couldn’t produce five versions, so you defended the one you had. Constraint made choosers out of all of us.

Now ideas are abundant. Execution is fast. Everything reaches a decent level almost automatically. AI removes friction, removes struggle, removes the cost of exploration.

What it cannot remove is the moment where someone has to look at everything on the table and say: this one! Not because it tested better. Not because it’s objectively superior. But because I believe in it enough to stand behind it.

That moment is not a small thing.

It’s where direction gets set. Where taste stops being passive and becomes action. Where work stops being a collection of options and becomes something with a point of view.

And AI cannot do it. Not because it isn’t intelligent enough. Because it has no stake in the outcome. No conviction that “this one matters” more than that one. No skin in the game.

You can generate options at scale. You can’t automate caring about one of them.

So here is what’s quietly becoming the most valuable thing in an AI-saturated world. Not prompting. Not knowing which tools to use. Not even creativity in the traditional sense.

Choosing. Committing. Being willing to discard the good in service of the necessary.

The willingness to say: this one, not that one, and mean it.

That’s not a skill AI is replacing. It’s a skill AI is revealing. Because until everything became easy to make, we didn’t notice how rare it was to find someone willing to make a real decision about it.

That’s the rearrangement nobody warned us about.Not that AI would take our edge. That it would show us how few of us had one to begin with.

This week we celebrated World Creativity and Innovation Day. Founded by the United Nations to focus on how creativity and innovation can drive impact across a wide range of the Sustainable Development Goals. The date falls six days after Leonardo da Vinci’s birthday and the day before Earth Day. Not by coincidence: It places human creativity between its origins and its obligations. In the space between imagination and responsibility.

Those of us who work at the intersection of creativity, media and technology should not let this day pass quietly. We are, whether we claim the title or not, the industry most responsible for proving that creativity drives real business value. That comes with a responsibility to stand up for it.

The business case for creativity is not creative

It starts with the consumer. The average attention span has fallen from 150 to 47 seconds over the past twenty years. The customer journey has shortened by nearly 40 per cent in just five years. Brands have less time, fewer touchpoints and an ever-narrowing window to capture attention.

Then comes AI. The volume of marketing content being produced is rising sharply, driven by generative tools that make production faster and cheaper. The result is a market flooded with content that increasingly looks and sounds alike. In a world of infinite content, sameness is the most expensive mistake a brand can make. And this dynamic will only accelerate.

The financial stakes are well documented. WPP ‘s “How Humans Decide” research, developed in collaboration with Saïd Business School, University of Oxford and drawing on WPP Momentum data, shows that 81% of sales are driven by brand priming effects built over time, not by in-market activation. Several others sources confirm this. Strong brands achieve market valuations up to four times higher than weaker ones. The cumulative value of the world’s most valuable brands grew 3.4 times between 1999 and 2024, reaching $3.5 trillion. These are not intangible outcomes. They are balance sheet reality.

And yet WPP’s Brand Asset Valuator (BAV), which has tracked more than 60,000 brands across 50 countries for over 30 years, shows that the number of strongly differentiated brands has halved in a decade. We are creating less of what truly moves the needle, precisely at the moment when it matters more than ever.

It will not be the biggest budgets or the latest technology that reverses this trend. It will be those who use creativity to build genuine relevance: a clear point of view, a cultural position, a real reason for people to care. This is where creativity does its most important work. Not as decoration or aesthetics, but as the core that connects people and brands.

Creativity is the path to imagine a better future

There is a larger point here. And it feels more important than ever.

We live in a time defined by uncertainty. That is almost a cliché by now. But it is true. Geopolitical turbulence. Economic volatility. A pace of technological change that outstrips our ability to keep up. The temptation is real: Feed the data into the model and let the system plan. And even create.

The problem is that historic data describes the world that was. Language models, however sophisticated, generate the most probable continuation of what has already been said or done. Neither produces the idea that has not yet been imagined.

Every major civilisational advance was not a projection from existing patterns. It was a departure from them. The printing press. Medicine innovation. The internet. None were outputs of extrapolation. They were outputs of human beings who looked at the world as it was and chose to imagine it differently.

We are at one of those inflection points again. The challenges ahead, from institutional trust in freefall to the societal consequences of AI itself, will not be solved by optimisation. They will be solved by people who combine deep insight with the courage to create something genuinely new and valuable. That is what creativity and innovation actually mean: not the aesthetic polish of a campaign, but the fundamental human capacity to build the future rather than simply fast-pacing it.

A builder’s perspective

What keeps me in this industry is a drive to design and build things. Not just to assemble components, but to see the whole before it exists. To draw the lines that others will eventually follow. Brands, strategies, teams, campaigns. Even, privately, Lego. Lots of Lego.

That instinct has always lived at the intersection of deep human insight and the act of creation. Understanding what moves people, and then designing something in response to it. Technology and AI have supercharged the capacity to build. But they have not changed where the ideas comes from. Human intelligence still draws the lines. Human judgment still decides what is worth building in the first place.

The organisations that will build the future are not those that use AI to replace human judgement. They are those that use it to give their people more space to exercise it, with greater depth and more courage.

We need to keep building the future. With technology, AI and ambition, but also remember not to be blinded by the lights, the speed and the glitter of progress and new stuff. We need to remember that unlocking the real value, comes from the human design, imagination and insight behind it. But this balance doesn’t happen by chance. It requires us to actively defend creativity and innovation.

And to remember what cannot be automated.