Every week there is a new model announcement, a new capability demo, a new post about how AI will transform everything. And every week there is the counter narrative: the failures, the job losses, the warnings that we are sleepwalking into something dangerous.
I live in this every day. I build with these systems. I see what they can do and I see where they fall apart. My biggest challenge in most meetings and conversations right now is holding both realities at the same time: this technology is genuinely extraordinary, and the transition it creates will be genuinely uncomfortable. Both are true. That is the part almost nobody is writing about.
Most people pick a side and argue from there. I understand why. Holding two contradictory truths at the same time is not easy. But picking a side is how you get blindsided, either by the opportunity you dismissed or the disruption you ignored.
This piece does not pick a side. It starts with what is already real, moves through the mess that is coming, and tries to be honest about all of it.
This is my actual morning
What I am about to describe is not a prediction.
At 8am, a WhatsApp message is waiting for me. Not from a colleague. From an agent I built called Neo. While I slept, it pulled my Whoop recovery data and assessed whether my body is ready for the crossfit workout thats planned. It checked the weather. It read my calendar and flagged a double booking at 11:30. It scanned my inbox and surfaced two emails that need attention, ignoring the rest. It reviewed yesterday’s meeting transcripts, extracted action items, and identified a commercial opportunity I had not explicitly discussed but that was sitting in the subtext of a client conversation. It cross referenced my open projects and flagged a blind spot where two workstreams are quietly diverging.
The message ends with my top three priorities for the day. Not the ones I set last week. The ones that actually matter this morning, given everything that has changed since.
I read it on my phone. I ask a follow up. I make a decision. The system acts on it.
Elapsed time: four minutes.
Let me be direct about what this is. I have AI employees. Not metaphorically. Not as a thought experiment. I have intelligent systems that are fast becoming as capable as people across a whole range of complex tasks. They do not sleep. They do not forget. They do not lose context between conversations. And they cost a fraction of what a human team would cost to do the same work.
Throughout the day Neo monitors silently and only interrupts when something genuinely needs my attention. It writes code, builds websites, with permission can send emails on my behalf, organises meetings and manages conflicts and focus, spawns cheaper models to handle research and drafting, and maintains a persistent memory of my projects, relationships, and priorities that compounds over time.
And it breaks. Regularly.
Right now, my Whoop integration is down because an authentication token expired. My web search capability is dead because an API key needs replacing. I have a second agent instance planned for my team that is waiting on a two minute setup task I have not got around to. A transcript to action item pipeline I need has not been built yet.
This is not a polished product. It is closer to a prototype than a platform. Some days it is extraordinary. Some days it is frustrating. Most days it is both.
But here is what matters. Even in this rough, broken state, it does something no software I have used in thirty years has ever done. It thinks about my work when I am not thinking about it. It connects information across systems that were never designed to talk to each other. It surfaces patterns I would not have seen for weeks.
When I sit down in the morning, I am already making decisions. Not gathering the information I need to make them. That difference sounds small. It changes everything.
This is the signal. Not the destination.
What I have today, cobbled together and fragile, is a glimpse of where OpenAI, Anthropic, Google, Meta and dozens of startups are taking all of us. They are building toward a world where this kind of intelligence is not an early adopter experiment. It is the default. And once it becomes the default, not having it will be like not having a smartphone in 2012. You could still function. You just could not keep up. However the implications are now a lot more significant to our livelihood!
Now imagine this applied to your team. Your sales pipeline. Your client operations. Your finance function. Your supply chain. Not as a single tool bolted onto an existing process, but as an intelligence layer that sits across all of them, connecting context, surfacing decisions, and acting on the things that used to require three meetings and a spreadsheet. That is not a five year roadmap. That is what is being built right now.
It’s improving hour by hour, day by day. The worlds biggest technology companies are building and deploying these at a speed never been seen before.
What this reveals about everything else
Here is the thing that using an agent like this forces you to confront, and it is more uncomfortable than the technology itself.
When the operational noise is removed, you see with painful clarity how much of your working life was consumed by it. Not as an occasional inconvenience. As the majority of the day.
Most of us never noticed because we had no point of comparison. You cannot see the water when you are the fish.
Be honest about what your week actually looks like. You open applications. You check dashboards. You attend meetings to relay information from one group to another. You compile reports so that someone senior can understand what is happening. You navigate internal systems, chase follow ups, update tools so that other tools stay accurate. You do this every week. You have done it for years. And you have called it your job.
It was your job. But it was never the point of your job.
It was compensation. The human cost of systems that were not yet intelligent enough to connect intention to outcome on their own. We built entire careers around being the bridge between what a company intends and what its technology can deliver. We mistook the bridge for the destination.
The applications you open every morning. Temporary. The dashboards you check. Temporary. The meetings where you translate between teams. Temporary. The career you built around being the person who knows how to navigate the complexity. Temporary.
Not temporary like a project that ends. Temporary like scaffolding. Necessary during construction. Invisible once the building stands.
The question this moment forces is not whether that scaffolding will eventually come down. It is whether you have started thinking about what you do once it does.
Why this time is different
I know what you are thinking. You have heard this before. Cloud was supposed to change everything. SaaS was supposed to simplify work. RPA was supposed to eliminate repetitive tasks. Every wave promised transformation and every wave was absorbed by the existing structures without fundamentally changing them. You have earned your scepticism. It is reasonable.
But every one of those technologies hit the same wall. They could handle structured, predictable, rule based tasks. The moment anything required judgment, context, interpretation, or reasoning across messy and incomplete information, the system stopped and a human took over.
That wall is the reason corporate structures survived every previous wave. The messy middle of organisational life, the part where someone reads between the lines of a client email, reconciles contradictory data from three different systems, navigates the politics of who needs to know what and when, that work was irreducibly human. Not because it was the highest use of human intelligence, but because no machine could do it.
Language models crossed that threshold.
Not perfectly. Not reliably in every case. But well enough, and improving at a rate that makes the trajectory unmistakable. For the first time in the history of computing, machines can reason across unstructured context. They can read a client email, connect it to CRM data, cross reference it with recent meeting notes, factor in the strategic priority you set last quarter, and surface an insight that would have taken a human analyst days to assemble.
This is not theoretical. OpenClaw, Manus, Claude with computer use, Devin, and dozens of agentic frameworks are demonstrating this today. They are rough. They make mistakes. They require oversight. But they are doing the connective, contextual, reasoning work that was supposed to be permanently human.
The wall that protected every previous corporate structure from technological disruption has a crack in it. And the crack is widening fast.
The mess in the middle
Here is the part that most articles leave out because it does not fit the narrative arc of smooth transformation.
The next three to five years will be extraordinarily messy.
Early agent deployments will fail publicly and spectacularly. A system will send the wrong communication to the wrong stakeholder at the wrong moment. An autonomous workflow will execute a decision that costs real money. A company that moved too fast will suffer a reputational disaster that becomes a cautionary tale on every conference stage for the next two years.
These failures will be real. They will cause genuine harm. And they will provide ammunition to every person who has been waiting to say I told you so.
The sceptics will have a golden period. Probably now until 2028. Every failure will be amplified. Every job displacement story will dominate the news cycle. The gap between what the technology promises and what it delivers in practice will dominate the discourse.
And underneath all of that noise, the technology will keep getting better.
This is the pattern that matters. Not the failures themselves, but the rate of improvement between failures. The system that sent the wrong email in 2026 will not make that mistake in 2027. Each failure will be specific and visible. Each improvement will be gradual and invisible. The sceptics will point to the failures. The trajectory will point somewhere else entirely.
The mess is real. The direction is also real. Holding both of those truths at the same time is the only honest posture.
The economics that make it inevitable
Even if you are unmoved by the technology argument, the commercial incentives tell you everything you need to know.
Every major SaaS company charges per seat. This model has generated trillions of dollars in enterprise value over the past two decades.
AI agents do not need seats.
An agent that can access the same system, perform the same analysis, and execute the same workflow continuously does not require a per seat licence. It requires compute. And compute is getting cheaper at a rate that makes the economic comparison brutal.
This comparison is already happening in boardrooms. Not as a philosophical discussion about the future of work. As a line item conversation about margins and competitiveness.
When a competitor demonstrates they can serve the same market with 40 percent fewer people and faster delivery, every board in the industry has the same conversation. Not because they want to reduce headcount. Because the market will punish them if they do not.
This is not about technology enthusiasm. It is about the structural economics of competition. Those economics are already in motion and they do not pause for consensus.
The costs of tokens per task. Displayed in my dashboard for all activity being delivered by an agent. Now compare that to your hourly rate? (Also check out humanortoken.com for full breakdown of cost per task.)
Trust, talent, and the questions nobody is asking
Here is a paradox the industry is not prepared for. As AI capability increases, public trust in AI will temporarily decrease. Every failure will be amplified. Every success will be treated as a threat. The more capable the technology becomes, the more threatening it feels to the people whose work it can now do.
The companies that get this right will earn trust through transparency: visible reasoning, legible decisions, meaningful human checkpoints. The companies that assume trust because the output looks good will discover how fragile that assumption is the first time something goes publicly wrong.
But the trust question is not the one that should keep you awake.
The talent question should.
Today, companies compete for people who can operate complex systems. The Salesforce administrator. The data analyst. The project manager who knows their way around Jira.
On the other side of this transition, that proficiency is worth almost nothing. The intelligence layer operates the tools. The human is valued for what the system cannot do.
This is a complete inversion of what “talent” means. The person who spent a decade mastering a platform becomes less differentiated overnight. The person who spent a decade building deep client relationships, developing ethical judgment, learning to read a room and make someone feel genuinely heard, that person becomes the most valuable person in the building.
Most organisations have not caught up with this. They are still training people to be better operators of systems that will not require human operators. Still promoting people for efficiency in tasks that will not exist. Still building career ladders with rungs that are quietly being removed from the bottom.
And underneath all of this sits a governance question that almost nobody is confronting. When your company’s intelligence layer knows more about your operations than any individual human does, who actually runs the company? If the system recommends a direction based on analysis no human can fully reproduce, where does accountability sit?
The answer has to be the human. Always. But for that to mean something, people have to genuinely understand what they are approving. Not just trust that it looks right. That distinction will define the difference between companies that thrive and companies that sleepwalk into a crisis they cannot explain.
The small company advantage
There is a widely held assumption that this transition benefits large enterprises with deep pockets.
The opposite is true. And this should unsettle anyone running a large organisation far more than the technology itself.
A twelve person company with intelligent systems can now operate with the analytical depth, the operational coordination, and the market awareness of a company ten times its size. No legacy systems. No middle management. No alignment meetings. No reporting cadences. Just a small group of highly capable humans directing intelligent systems toward outcomes.
The competitive threat is not that other large companies will adopt AI and become more efficient. It is that small companies will enter your market with a fundamentally different operating model and a cost base that makes your structure look like a liability.
The advantage of scale used to be coordination. When coordination becomes automated, the advantage of scale evaporates. What remains is brand, relationships, and proprietary data.
If that is all that remains, you had better be sure yours are strong enough to hold.
The mirror
Here is the exercise that will tell you everything you need to know.
Write down everything you did last week. Every task, every meeting, every email, every report, every piece of admin. Be honest about it. Do not editorialise. Just list it.
Now draw a line through everything that was essentially moving information from one place to another. Compiling data someone else needed. Attending a meeting to relay something you already knew. Updating a system so that another system could be accurate. Chasing a follow up that should have been automatic.
Look at what is left.
That is what your job becomes on the other side of this.
For some people, what is left is substantial. A rich core of judgment, creativity, relationship, and strategic thinking that was always there but buried under operational noise. For others, what is left is thin. Not because they lack talent, but because the role was built around being the bridge.
The honest question is not whether the bridge is coming down. It is what you are standing on when it does.
The work that actually matters
I know where this is heading because I can already feel the shape of it in my own morning. A WhatsApp message. A few minutes of thought. A decision. And then on to work that is unmistakably human.
A conversation with a client who needs someone who understands what they are going through. A strategic question with no right answer, only trade offs that require wisdom. A relationship built not because a system flagged an opportunity, but because I believe in what someone is trying to do and I want to be part of it.
That is the work that matters. It always was.
But here is the part I do not want to leave unsaid, because it is the most practical thing in this entire piece.
The smartest people I know, the ones who are always a step ahead, have already moved past using AI to write emails and generate content. They are building agents. They are taking the messy processes they do every day and turning them into systems that run alongside them. They are learning a new skill, and it is not prompting.
It is orchestration.
How agents work. How to connect them to your context, your data, your workflows. How to direct them the way you would direct a capable team member who needs clear objectives and regular checkpoints. Not coding. Not engineering. Thinking in systems. Breaking a process into components an agent can handle. Maintaining oversight without becoming the bottleneck.
Most people are still treating AI as a better search engine or a faster writing tool. That is like learning to type faster when the entire concept of typing is about to change.
The generative layer, the chatbots, the content creation, that is the visible surface. Underneath it, the agentic layer is where the real shift lives. The people who learn to work at that layer, who can take their domain expertise and direct intelligent systems toward outcomes, will have capabilities that are genuinely transformative. And the people who are naturally good at this are often the ones you would least expect: generalists, problem solvers, people who see across silos and understand context. If that sounds like you, you are closer to this than you think.
The scaffolding is coming down. Not cleanly. Not painlessly. Not without real cost to real people along the way.
But here is what I know from living inside this every day. The work that matters, the judgment, the creativity, the relationships, the ability to sit with another human being and actually help them, that work is not going anywhere. It is the only work that was ever truly yours.
Everything else was borrowed time.
The question is whether you see that now, while there is still time to act on it, or whether you see it later, when the choice has been made for you.
If any of this resonates and you are not sure where to start, reach out. I am happy to point you in the right direction.
If this resonated, subscribe. The next piece explores the companies that are already operating this way, what they look like from the inside, and why the transition is not as far away as most people assume.
Craig Hepburn is an AI strategist and builder, Perplexity Fellow, and former Chief Digital Officer at Art Basel and UEFA. He works across technology, business, and system design to shape how AI operates responsibly in the real world.
Great reflection Craig! I've got a Mac Mini on the way...not sure if I will go with OpenClaw or PicoClaw, but I agree this is important and something worth investing in to stay ahead of the curve and identify where there could be valuable new kinds of business.
This is exactly my sentiment too - using AI to deal with the noise so humans can be less stressed and more productive. It's my headline narrative in https://www.maxy.bot/.