We've Been Asking the Wrong Question About AI
After two years of intensive work with generative AI, stepping away from our day-to-day digital transformation consulting to understand what these systems can actually do properly, I've arrived at some conclusions that should change how we think about AI adoption in small businesses. Not the hype, not the fear-mongering, but the messy, practical reality of making this technology work for independent operators.
The LLMs have got exponentially better in this time. Whether we'll see huge step changes from here is unlikely; however, even with the limitations of the underlying technology, they are getting more impressive.
That final transformational leap will come from a different kind of model entirely. But here's the thing: even as they stand, these systems are transformative. We just keep asking the wrong questions about how to use them.
The Mission Statement That Should Scare You
Let me read you something from OpenAI's mission statement:
"AGI, by which we mean highly autonomous systems that outperform humans at most economically valuable work."
Read that again. Highly autonomous systems that outperform humans at most economically valuable work. The North Star focus of this technology is to do what many of us are paid to do.
Now, before you dismiss this as dystopian scaremongering, let's be clear: we've empirically proved in the last 12 months that there are lots of jobs - particularly in the events and B2B sector - that AI can already do better.
And AI doesn't stop. It doesn't need a tea break. It doesn't need to sleep. It keeps going. When a business person looks at that reality, why wouldn't they pursue it?
The CEOs can say they're worried about job displacement all they want. Look at what people do, not what they say. They know it's going to take jobs. They're working in a construct where huge macroeconomic forces mean they can't simply stop, even if they wanted to. They've got shareholders, investments, a whole freight train of people pushing them forward.
But we can think about it differently and make it work to our advantage.
A Word of Warning: Stop Shooting Yourself in the Foot
Here's something I see constantly, and if you're doing this, please stop. Stop sending pure AI-generated work to your boss. The only signal you're giving is: AI can do my work. Which may or may not be true, but your boss - no matter what they say - will look at that and start seeing an AI opportunity where your job used to be.
Do yourself a favour. Look around the internet and learn how to use these tools effectively. Because the question isn't whether AI can replace you. The question is: how do you become indispensable by using AI better than anyone else?
We've Been Asking the Wrong Question
Where do humans fit into this AI-centric world? We've been approaching this from entirely the wrong angle.
Let's step back and look at knowledge work. There are huge inefficiencies in knowledge work - inefficiencies that stem from our tooling. Yes, we've digitised things, but we haven't really evolved significantly from the days of paper and filing cabinets. The efficiency gains in knowledge work are nothing compared to what's happened in manufacturing and blue-collar jobs.
Manufacturing workers have been "optimised" for decades. They've arrived at a position where humans and automation are symbyotically deployed where each adds the most value. And here's the key insight: we've seen 50x productivity improvements in manufacturing. Knowledge work hasn't even come close.
The reason AI is going to take many jobs is precisely because of this lack of productivity in knowledge work. We've just never had the tools to achieve that 50x leap. And I believe LLMs - combined with other technologies - are fundamentally that tool. They unlock our ability to reach those productivity levels.
The Small Business Productivity Trap
I've worked in startups and small businesses all my life. When there's a problem with productivity or a bottleneck, what's the typical response?
- Hire an intern
- Create a management structure (resulting in a useless layer of bureaucracy)
- Buy work management software
Suddenly, you've got five different systems doing different things. Everybody's busy updating systems. None of them is actually up to date. And then those of you who started the business go down the pub and look at each other with that familiar lament: "We got more stuff done when there were three of us. What on earth is going wrong?"
This is where the opportunity of this technology becomes clear.
A Real-World Example of Knowledge Work Inefficiency
Here's a scenario that actually happened (anonymised, but absolutely real): Marketing wants to know how the tier-one audience has grown over the last three years. You now have a data team, so it gets sent over.
The data team conducts analysis. Three days later, it gets sent back - probably via dashboard. Marketing says: "This can't be right."
Emails fly back and forth. "How do you define tier-one audience?"
"Well, we define it like this."
"That's not the correct definition. Use this one."
Comes back.
Results look even worse. What does marketing do? They spin up Excel, fix it themselves, and put that forward.
Then someone suggests: "We need a data literacy programme and an AI system!" (Because AI solves everything.) You end up with a costly AI system that requires extensive training. There's a podcast where one of my contemporaries said
"If your system needs too much training, it's not going to get adopted. Don't buy it."
Absolutely true. There will be too much change management, adding more inertia than you already have whilst introducing tooling whose job-displacement potential many will view with distrust.
As a bonus, you get a data literacy programme that's half-baked, stuck in the eighties, completely useless, and wastes everyone's time.
Now Invert Your Thinking
What if we approach this differently? What if we start with: "I have LLMs. What can I do with them?"
Same scenario: Marketing team needs to answer that question about tier-one audience growth. They open Claude or ChatGPT Teams. They type in the question. They get back the answer. They can iterate back and forth right there. It all works.
The data team isn't getting frustrated. They're not being asked to rejig dashboards because someone doesn't like a pie chart and wants a bar graph instead. All that back-and-forth is gone.
Here's the secret reality: That scenario I just described - having a conversation with an LLM to get that answer - you can do it now. It doesn't actually require lots of systems. The SaaS companies are scared of you knowing this, because it's entirely possible to do.
We're going to be doing this on Digitising Events and sharing it with you.
Human in the Loop: What It Actually Means
This brings me to a concept I call Human in the Loop (HITL). And by Human in the Loop, I don't mean: "Human gets an email saying, 'Here's some generative AI output. Do you like it? Yes/No?'" That's not what I mean at all.
What I mean is putting people into an AI-centric, agentic process - where we enjoy what we do and create way more value than we could if we weren't in it. My inspiration for Human in the Loop comes from "driver in the loop" in racing car simulators. You are completely encased by the technology. It makes you feel like you're part of the action. It is real. You get responses and immediate feedback.
That is Human in the Loop. That's how AI process design should work.
Beyond the LED Bulb Moment
Here's an analogy from media businesses I've run in the past, which were in the energy efficiency sector. When energy efficiency became a thing, everybody talked about: "Replace your bulbs with LED lighting!" And if you're a huge factory, that gives you an immediate hit.
That's how I see the current AI adoption landscape. Everybody's giving their employees access to GPTs (whatever flavour you choose), and they're seeing a hit. There's a great study showing that, on average, each employee regains 1 to 2 hours.
But that to me is the low-hanging fruit. That's like switching to LED bulbs.
If you really want to tap into the potential, you need to go out and find the opportunities others can't see.
In terms of AI utilisation - and I'm not talking about the hugely clever people building these models, I'm talking about the practical application of AI - we're all stuck somewhere between the plate and first base. None of us has really hit first base yet. It's a huge, huge opportunity.
Finding these opportunities is challenging, and most companies will back out of it - especially the big ones. If you're a small entrepreneurial company, that is your superpower. You can make those changes and apply them now.
Seven SaaS Platforms Eliminated
By applying this thinking, we have eliminated seven SaaS platforms from our business, whilst doing the work we want to do better, with less friction. Seven.
That is all about thinking: How and where do you now put these agents? What do they do? How do you reimagine the way your organisation produces value?
Let me share some examples.
Content Generation: Beyond the Asinine Promise
Remember the promise from 24 months ago? "Type in a few sentences and you get a blog post!" Tried and tested. Failed.
What we've done is different. We've looked at content intelligence and asked: Can we use AI to generate a "further reading" section that is written in context with the article it's being displayed on? Can we write the intro to lead people into those suggested readings? Can we write the description of each recommendation in context, so it's a bit like a human saying: "This article is good, but so is this one because..."
That's now accessible. We can do it. The one thing we're working on is: when you click on it, how can we take you to the exact point in that user journey that matters?
Why is that important, other than sounding clever? Because our engagement time has gone through the roof. People are reading and clicking through because they're being guided.
The old world was an editor looking at related articles and saying, "Hey, I think these three." You'd have those three links with the same generic titles, and you're asking people to take a leap of imagination to decide whether they're relevant. What we're doing is explicitly explaining why it's relevant, so they actually read it.
It is about reimagining how we present and personalise content. And what I've described is scarily simple. The big platforms must be really worried about how simple this has become.
Data-Driven Content Strategy
We've now freed ourselves up to really look at the data and ask: Where are the clicks? Where's the engagement? How do we feed that information back into our content process design so the next piece of content we create aligns with the value the audience actually wants?
For the first time, we have a toolset that can operate platforms for us.
If you haven't already, look at the concept of MCPs (Model Context Protocol), where you can control updates and platform changes in a remarkably simple way. It's only remarkably simple because we have these LLMs with such immense power, just waiting to be unlocked.
The Project Management Win
The biggest win I see is in project management. You get whatever tool you want - Asana, Monday, whatever - and you're updating stuff like crazy, yet you still don't know what's going on, particularly in a small business where you're changing things quite fluidly (which is both the fun and frustrating part of small businesses).
How can we put a layer of intelligence around a small business so you know what's going on without having to update loads of different systems? You just know. Imagine what that would do. That would be powerful. We could just get on and do what we need to do.
The Anthropology of Work
Here's an anthropological observation: we work most effectively when we're not distracted and focused on a single outcome. Think about all the systems we put into work - they're all needy little things vying for attention.
Ping: You've got an email. Ping: You've got a message. The more disciplined among us turn these off, but they are fundamentally attention-seeking.
I think there's an opportunity to create something that just knows. It doesn't need attention. It'll prompt you at the right time when you're not deep in thought.
What We're Doing with Digitising Events
Here's what we intend to do with our AI adoption:
We will experiment in full view of the world on Digitising Events. We will explore different ways to apply AI. We'll share them with you. We'll share the results. You can adopt them or not - it's your choice.
But we're going to do it in a Human in the Loop way. Every time we share something, we'll explain how it'll help your talent be better.
Most importantly, I think Human in the Loop is the only ethically responsible way of adopting AI. We should always ask: "How do we make our people better?" Rather than: "How many people can this replace?" That's a slippery slope, and we should try to avoid it.
The Opportunity for Small Businesses
We're quite excited about this scary and exciting journey we've chosen. The reality is this: large companies will struggle to make the operational changes required to truly leverage AI. They'll buy expensive systems that require extensive training. They'll implement data literacy programmes that waste time. They'll create more bureaucracy.
Small entrepreneurial businesses don't have those constraints. You can reimagine workflows today. You can eliminate SaaS platforms tomorrow. You can build AI-centric processes that make your team more effective without the organisational inertia that plagues larger companies.
It is your competitive advantage. This is how independent operators can outperform big rivals.
We'll share all the tools we use and our views on them. Hopefully, we can help all of us, entrepreneurial small businesses, navigate this AI revolution together.
That'd be fun.
