Have We Met Before…?
For anyone who builds the systems people are asked to trust.
Oscar Wilde wrote: “Experience is simply the name we give our mistakes.”
He shared that in 1892. Three years before Marconi transmitted the first radio signal across a field in Bologna. A century before the browser. A century and a quarter before the large language model.
He was talking about people, generally. But he might as well have been talking about us. Because this is a piece about experience. The kind we have already had. The kind we appear, collectively, to be ignoring.
The Shape of the Promise
Consider this, if you would.
“A new medium is emerging. It will connect every person on the planet. It will democratise access to knowledge. Flatten hierarchies. Give everyone a voice. It will transform education. Commerce. Healthcare. Government. Distance will become irrelevant. Information will be free. Human potential, finally, will be unlocked.”
That is the internet. Mid-nineties. The promise was everywhere: in op-eds, in congressional testimony, in the kind of evangelical certainty that only arrives when a technology is new enough that nobody has lived with its consequences yet.
Now, how does that compare to this?
“A new technology is emerging. It will democratise access to expertise. It will augment human capability. It will transform education, healthcare, creative work, scientific research. It will give everyone access to tools that only the privileged could afford before. Human potential, finally, will be unlocked.”
That is the AI pitch. 2023.
Same shape. Different noun.
Much of the first promise came true, in some form. The internet did connect much of the planet, unevenly, imperfectly, but at a scale previously unimaginable. It did open access to knowledge for many who had never had it before. It did loosen certain hierarchies and give certain people a voice they previously lacked. The problem was not the promise. The problem was what we chose to optimise for, first.
The internet was and still is a marvel. Imagine what it could have been, if the wisdom had arrived before the business model did.
The past does not predict the future. But it does illuminate it. Enough to navigate. Enough to govern. Enough to design with intention rather than hope.
The Wrong Order
The fast things came first. The important things were scheduled for later. And later kept getting pushed.
That is the entire story of modern tech. And it is worth understanding precisely how it happened, because the sequence is running again.
Connection arrived first. The ability to reach almost anyone, almost anywhere: slow, unreliable, but remarkable in a way that is easy to forget now. For many people, for the first time, publishing was possible without a publisher. Finding someone on the other side of the world was possible without an institution behind you. Because connection came first, it shaped the conditions for everything that followed.
Then the business model arrived. Attention, harvested and sold to the highest bidder. Not designed for humans. Designed for growth. It optimised for engagement, and it turned out that outrage, fear, and compulsion were far more engaging than understanding.
Notifications learned your name before your friends did. Not your actual name: your behavioural signature. What time of day you were most anxious. What kind of headline made you tap. What kept you scrolling at midnight when you should have been sleeping. Your friends took years to understand you. The algorithm had a working model in weeks. It did not know you. It knew your weaknesses. There is a difference.
We built infinite scroll before we understood what it did to sleep. Sometime around 2016, we started properly noticing what we had built. The voice for everyone had, in too many places, become a megaphone for the loudest and angriest. By which point it was deeply embedded infrastructure for a significant portion of the planet, woven into how many elections were fought, how news travelled, how public opinion formed.
The wisdom, the governance, the serious thinking about second-order effects, came last. Or has still not fully arrived.
Nobody designed it to go wrong. The engineers shipped. The philosophers were in the room next door. They were not invited to the sprint, because the sprint did not have a field for consequences.
And now, here we are again. Though I want to be careful about that phrase, because the parallel is not perfectly clean.
The speed is here. The fluency is extraordinary. Capability arrived fast and it is compounding faster. The business model looks familiar: move fast, capture market, lock in users, extract value. Many of the companies building AI are the same companies that built the surveillance internet or are operating from the same incentive structures. They did not, for the most part, become fundamentally different kinds of organisations. They became larger ones.
The Overton window has shifted. Some labs are treating safety and governance more seriously upstream than early internet architects largely did. Red-teaming, model cards, institutional safety teams: these things exist now in ways they did not in 1995. That matters, and it would be unfair to ignore it.
What has not changed is the underlying incentive architecture. Regulation applied from outside, after deployment, is not the same as consequence built into the foundation before anything ships. The philosophers are still in the room next door. They are still not always invited to the sprint.
And there is something genuinely different this time that the parallel risks smoothing over. The internet automated distribution and coordination: human ideas and communication, moving faster. AI is beginning to automate cognition itself. Thinking. Judgement. Reasoning. The internet shaped what we paid attention to. AI will shape what we believe we know. The displacement effects on expertise, education, and economic value may be faster and more structural than anything the internet produced. What that means at scale, I am not certain. I am not sure anyone is yet. But it deserves sitting with, rather than setting aside in the rush to draw the comparison.
Different building this time. Same room. Bigger fire.
What I Have Seen
I have spent nearly thirty years inside these systems. Aviation. Energy. Financial services. Payments. Pharma. Retail. Fashion. Telco, and many more. Each one has its own relationship with failure, and its own version of what happens when the consequences of getting the order wrong are not measured in engagement metrics.
Some are more consequential than others. In aviation and energy, a hallucination is not a funny tweet. It is a catastrophe. Potentially fatal. I am not speaking from the outside looking in.
We’re all always learning… and there is something I learned the hard way, at one of the world’s largest global banks: an institution managing trillions in assets across more than sixty countries, with hundreds of millions of customers and a design estate to match.
I walked in and started measuring against the existing operating model. The existing org structure. Get the analytics in. Show the numbers. Demonstrate the gap. Present the solution.
What I learned, slowly and not without real pain, is that measuring the wrong things with great precision is just expensive confusion. And you cannot walk into an ecosystem that has sustained itself on stale bread and force-feed it pizza. They will not like it. They do not want pizza. They do not believe in pizza. You have to let them smell it first. You have to demonstrate what different looks like before anyone will consider changing what familiar feels like.
Transformation is not, in my experience, primarily a logical problem. It is a sensory one. People rarely change because they are shown a better argument. They change because they are shown a better reality: close enough to smell, low enough stakes to consider. The argument comes later, once the smell has done its work.
That is what The Experience Academy, a vehicle I created then let run organically within the teams under my remit, was really about. Not reporting. Not governance theatre. Letting the organisation smell the pizza. Creating the conditions in which a different kind of work became conceivable, before asking anyone to commit to it.
The fundamentals of what became F.R.A.M.E.™ (my Ai supported governance backed design thinking protocol) were identified inside that global bank. Developed further at a major international commodities intelligence and data business serving the energy markets. Then tested at real scale inside one of the world’s largest integrated energy companies, where I led a global design capability across more than forty fragmented systems, eight major digital platforms, and a team spread across multiple geographies.
Three years into my six years at that energy business, we got a new CEO. If you have ever worked in a large organisation, you know what that means. It is not just a personnel change. It is a signal. Ambiguity breeds anxiety. The question that runs through every corridor, every one-to-one, every Friday afternoon: what does this mean for us?
I sat my entire capability down on a global conference call. Two hundred people. And I told them: change is coming. We do not know what kind, or when, or exactly how it will affect us. But it is coming. Be prepared.
The reason we did not have to scramble is because the shelter was already built. The case studies, the commercial evidence, the documented impact: baked into the personal appraisal process before anyone needed it. Every practitioner had a body of evidence showing what they had done, why, and what it had delivered. You do not build the shelter when the storm arrives. You build it before.
The consequences had a return on investment. Forty-plus fragmented systems became one coherent platform. Broken governance became clear mandate. The work became commercially undeniable: protecting value, reducing risk, contributing to tens of billions in revenue and tens of millions in operational savings.
But the numbers are what smiling at scale looks like on a spreadsheet. Good governance is not bureaucracy. It is how you care for people at scale. It protects them from burnout, from bad decisions made in ambiguity, from moral injury, from being the last to know when something goes wrong.
We are nothing if we do not look after our people.
The Most Expensive Lesson I Have Witnessed Recently
Since leaving that energy business I rebooted my consultancy, Experience Artisan.
I would like to share something about one organisation, whose name I will not use, but the pattern is familiar across multiple engagements I have been having.
They ran a pilot. Proved the concept. Then went hell for leather, deploying across the whole organisation and making a fanfare of it. We are an AI organisation now. This is the future.
Then the handbrake turn.
Redundancies. Restructures. Good people let go because the numbers were supposed to work differently. The operational wins never came. The customers felt it before the leadership did. The leaders who survived were exhausted. The ones who left were, in several cases, genuinely damaged.
There was a moment, and there is always a moment, when someone in that organisation saw the gap between what the pilot had proven and what full deployment was requiring. The systems were not ready. The operating model had not changed. The people had not been brought along. Something was flagged. A meeting was held. A decision was deferred. That moment passed without intervention, because there was no governance structure that made intervention structurally possible. The fanfare had already been made.
Where was the governance that should have asked, before a single deployment decision, what happens to the humans in this system if it goes wrong?
They lost it all. And now they are rebuilding from a position of damaged trust that a proper foundation would have protected them from.
The lesson is simpler than anything they did.
Understand what you want to be as an organisation. Apply an operating model to support that. Then, and only then, deploy AI where it genuinely serves the people inside the system and the customers it is built for.
Fix the house first, before you start redecorating it with AI.
For those building at the frontier rather than adopting into existing organisations, this is not advice that only applies from the outside. The organisations racing fastest are still organisations. They still have sprint cycles, incentive structures, and the same tendency to schedule the important things for later. A deployment fanfare followed by a redundancy announcement is not only an incumbent pathology.
It is a human one.
Where It Actually Lives
There is a version of this argument that positions designers as the missing adults in the room: the people who, if only given proper authority, would fix what engineering and product and executive leadership have broken. I have been in enough rooms to know that is not how it works.
The core issues around capability, alignment, emergent behaviour, and misuse sit deeper than the interface layer, in model training, data curation, evaluation, and the fundamental science. Good interface design cannot fix a model that confidently hallucinates in high-stakes domains.
What I am arguing is something more specific. The interface layer compounds the technical layer in both directions. Even a well-aligned model can be made epistemically dangerous by a product layer that strips out its uncertainty signals and presents its outputs as authoritative. The model and the interface are not separate problems. They interact. And right now, that interaction is being managed with insufficient seriousness.
Interfaces have historically been honest about their own limitations. A progress bar says: I am working. I am not finished. I might be wrong. Interfaces expose state. They show seams. They carry an acknowledgement that something is happening underneath, and that it could fail.
There is largely no such signal in most AI products. No visible uncertainty. No confidence interval. Just a clean paragraph and a blinking cursor, waiting for your next question. A system that is forty percent confident is presented with identical typographic authority to one that is certain. Same font. Same weight. Same spacing. No visible difference. Yet.
Typography is a trust signal. Using the same weight for a guess as for a fact is, in the most precise sense of the phrase, lying with CSS.
The model has signals of uncertainty. We chose not to show them. That was a choice. We made it. If the system looks certain, the user behaves as if it is. Fluency mimics understanding. A system designed never to say I do not know is among the more consequential things we could put into the hands of people making decisions that matter. We built something close to that. We shipped it. We called it seamless.
Some teams are already working differently: designing for uncertainty, building in verification loops, surfacing explicit reasoning traces. They are not yet the norm. They are the exception. But they prove it is possible. Showing seams is not friction. It is honesty infrastructure.
The internet, in the way it was ultimately built and monetised, largely exploited attention. AI, in the way it is currently being deployed, appears to be doing something similar with trust. The ethics of AI do not end in the model. They live inside the system. Every decision. Every layer. The incentives have not fundamentally changed. Only the technology has.
The Ability to Evangelise
There is something I wish someone had said to me earlier in my career, so I will say it here.
The work only counts if the right people know it happened.
Visibility is accountability. And evangelism is how visibility scales.
Not a slide deck. Not a quarterly update that nobody reads. The ability to stand in a room, any room, and tell the story of what the work did. What it saved. What it generated. What it protected. In the language of the person you are standing in front of.
With a CEO, it is revenue and resilience. With a CFO, it is cost and risk. With a team, it is purpose and craft. Same story. Different register. Always true.
Redesigning the Table
Getting a place at a table that was never designed to ask the right questions does not change what gets decided at it.
The governance structures around which design currently orbits were not built to handle what is coming. That is not a criticism of the people who built them; it is an observation about the moment we are in. The question I think worth asking is not how to earn more influence within those structures, but whether the structures themselves are adequate. In my experience, they are not.
Design needs to align with governance. But more than that, design needs to redesign the table. The structures through which consequence is evaluated. The mechanisms through which second-order effects are made visible before deployment rather than excavated afterwards. The frameworks through which the humans inside a system are protected before the storm rather than compensated after it.
What might that look like in practice? Confidence calibration surfaced at the product layer, not buried in documentation. Design accountability embedded in pre-deployment review, not bolted on post-launch. Governance as a design material: not a constraint applied from outside, but a foundation built from within.
Those structures are not fixed. They were designed. They can be redesigned. That is not a metaphor. That is a brief.
The Question
Some of you are reading this after a full day of tickets, stakeholder reviews, and someone asking you to make the button bigger. The last thing you need is another voice telling you the stakes are high.
The permission most people are waiting for is not coming from above. The organisations that put the right people in the room before the sprint starts, not after the damage is done, are the ones that will still be trusted when the reckoning comes.
That is not idealism. That is a pattern I have watched play out enough times to be reasonably confident in it.
When the regulator comes, and they will, the organisations that designed for trust, for governance, for consequence, will still be standing. The ones that designed for seamlessness at the cost of honesty will be explaining themselves.
I want to be clear about what this piece is and is not. It is an opening argument, not a complete account. Who decides the actual boundaries when capability outpaces institutions? Not the designer. Not the product team. Increasingly not even the organisation. That is a question about law, about economic incentives, about the relationship between nation-states and the handful of organisations controlling frontier capability. It deserves its own treatment.
This is the first question, not the last.
Oscar Wilde said: “Experience is the name we give our mistakes.” I have made enough of my own to know he was right. So I am saying:
Design is the name we give our intentions.
So here is the question to sit with. Not in the abstract. In the next thing you ship. In the decision you make tomorrow morning, before the first meeting starts.
What are you designing it not to do?
Because if you do not decide, the business model will.
It always does…