I’ve watched the same pattern repeat for twenty years now.

Internet adoption. Mobile. Cloud. Now AI.

Every time, organisations buy the tools, book the training, send people to learn the new thing. Then they wait for the transformation.

It doesn’t come.

The research backs this up: only 10-20% of training transfers into workplace practice. That means organisations waste 80-90% of their training investment when learning is disconnected from real work.

With AI, the failure rate is even starker. AI project failure rates range from 70% to 85%. Some research suggests 95% of AI initiatives fail to yield meaningful business results.

The most common reason? Misunderstandings about project purpose and domain context.

The Tool Isn’t the Problem

People ask me all the time: which AI tool should we use? Claude or ChatGPT? Gemini or Perplexity?

My answer: it really doesn’t matter.

You can do most things with most models. The tool is a small percentage of the equation. When you’re doing complicated work or very specific tasks, there might be something that works better. But for most use cases, the tool won’t matter much.

Get the process right. Organise the people. The rest falls into place.

Whether it’s Claude or ChatGPT probably won’t matter in 80% of cases.

What Actually Breaks AI Projects

I’ve seen a pattern in SMEs bringing us in because their AI implementation isn’t working.

The first thing I look for? Communication patterns. How teams talk to each other. Whether there are silos. How they share information, give feedback, discuss what’s working and what isn’t.

It’s organisational behaviour first. Not the technology.

The most common scenario: an SME leader worries that people are resisting AI. So they mandate use of AI. But they’re not giving direction or vision. They’re just telling people to do stuff quicker.

They’re anxious about adoption. They should be worried about coordination.

If everyone’s in silos and they start using AI, they’re still just using AI in silos.

Another common one: Microsoft rolls out Copilot to everyone. Ta-da. Access granted. Let’s go.

Nobody knows what to do with it. There’s no training—but it’s not just about training. There’s no cohesion. No vision. No strategy.

You can’t just throw tools at people and expect them to do stuff.

The real reason they’re not using it isn’t lack of training. When you start using these tools, you get into it. The problem is having no shared vision about what’s happening and where you’re going.

Generic Training Creates Awareness, Not Capability

Generic AI literacy training creates awareness. That’s it.

When AI training is too broad, learners check the box but fail to bridge the gap to their specific daily tasks.

An accountant needs to know how to use AI for anomaly detection. A marketer needs it for ideation. Not generic prompt engineering.

The fix: focus on the problem, not the technology.

This mirrors what RAND research found about AI project failures. Organisations that focus on using the latest technology rather than solving real problems for users see significantly higher failure rates.

AI projects require at least a year of commitment to a specific problem. If an AI project isn’t worth that long-term commitment, it’s probably not worth starting.

Sound familiar? That’s the apprenticeship model. Sustained, problem-focused, embedded over time.

How We Teach AI (and Why It Works)

Our apprenticeship delivery works differently.

At the beginning of each month, there’s a group session. People learn a new framework, tool, practice, skill, or way of looking at things. A model they could try out. A case study they could find a similar version of in their own work.

They practise this in a group.

Then they go back to work. They try to implement what they learned. Or they learn more about it whilst fixing things at work, making things better.

Towards the end of the month, everyone gathers again for group coaching. Peer-to-peer feedback. Everyone brings back what they did.

You get this peer learning thing chaired by a technical expert. You’re developing all the core skills that are necessary. The learning experience works because it’s enjoyable, you’re doing things at work, and it’s reducing resource strain on the team.

The things you’re doing are actual work. Making work better.

The time out of the office—the training in inverted commas—is minimised. The outputs are maximised.

The work becomes the curriculum.

Research shows that companies investing in embedded employee development see a 24% increase in productivity compared to traditional training that pulls people away from their work.

The Pattern That Keeps Repeating

Across internet adoption, mobile, cloud, and now AI, one pattern keeps showing up.

The connector skills matter most.

Communication. Collaboration. Being able to summarise. Give and receive feedback. Understand the wider context of the work you’re doing. Research and learn.

These generalist skills—the connector skills—go into all the technical disciplines we cover at The Coders Guild.

We’ve spent years researching how people learn. What works. What doesn’t.

It’s always those core skills. These generalist capabilities that we keep coming back to.

The technology changes. The human capabilities that determine success stay the same.

Why SMEs Have an Advantage

Here’s something that surprised me about how SMEs adopt new technology.

They’re able to pivot and deliver business change much faster than big organisations.

When you’ve got just a dozen people working in a department or team using a CRM or particular system, swapping it out for something else happens quickly. Delivering something valuable to that team is faster because you’ve got fewer people to talk to.

You get to talk to everybody involved. You can make better choices really quickly.

Business change is much faster in SMEs.

This contradicts the narrative that SMEs are at a disadvantage. They’re not. They can move faster and make better decisions because they can actually talk to everyone involved.

What Service Design Teaches Us About AI

My background in service design shapes how I approach AI work.

I start customer-first. When I’m looking at a service design process, I’m using UX principles. Easiest route to achieving the thing you want to achieve—or the user wants to achieve.

Customer-focused, not tech-focused.

We’re not thinking about the platform. We’re thinking about the journey. Reducing friction in a journey.

A pure tech consultant probably wouldn’t do that. They’d start with the platform, the architecture, the implementation.

I map journeys and reduce friction. Customer-first, not tech-first.

That’s what makes AI implementations work. Understanding the actual workflow. The real constraints. The people doing the work.

You Can’t Train Your Way Out of a Broken Process

If the problem is dirty data, bad workflows, or unclear approval processes, the solution is a systems fix.

Not a learning course.

AI training fails when it ignores the actual constraints in the work environment.

Traditional training suffers from the forgetting curve. Employees lose most of what they learn within a week if they don’t apply it immediately.

Learning embedded in the flow of work ensures knowledge is reinforced at the moment of need. That dramatically improves long-term retention.

Research on cognitive apprenticeship shows that teaching and learning are deeply embedded within complex and highly variable contexts. Making expert thinking visible—not just observable actions—is critical for knowledge work and technical roles.

What Works Instead

Context is the curriculum.

Learning AI whilst solving an actual business problem you care about. With real stakes. Over real time.

The apprenticeship model works because it embeds learning in real work. People learn frameworks, apply them to their actual problems, then share what worked and what didn’t with peers.

Even in year two, apprentices are already rated between 70-75% productivity of a traditional full-time trained employee. Research from Harvard shows that young people really can do the job if they’re given the right resources.

The pattern holds across every technical discipline we’ve delivered.

Software development. DevOps. Testing. UX. Accessibility. Agile project management. Now AI and automation.

The technology changes. The learning model that works stays the same.

Get the process right. Organise the people. Embed the learning in real work.

The rest falls into place.