I recently had a good exchange on LinkedIn with someone who had spent 37 years in construction. His argument was that, for decades, contractors have bought software designed by people who have never actually done the job. They've never been on site, dealt with permits, managed a delayed project or had to explain to a customer why something hasn't gone to plan.
His point was that AI is changing this. Someone with decades of industry experience can now use AI to start building software themselves, without necessarily needing the traditional software development background that would previously have been required. And I think that's genuinely exciting. We're already seeing people use AI to turn their own knowledge into tools, automation and software that might never have been commercially viable to build before.
But I made the point that the same argument works in reverse.
I've spent more than 25 years building software. That experience has taught me what works, what doesn't, where systems tend to fail and which shortcuts can become very expensive several years later. I've seen databases that worked beautifully with 100 customers grind to a halt with 10,000. I've seen seemingly harmless architectural decisions make systems incredibly difficult to change. I've seen security problems, failed integrations, data issues and systems without adequate backups or recovery processes.
That experience matters too.
The $10,000 tap with a hammer
There's an old story that's been retold in various forms about an experienced engineer who is called in to repair a large machine. Nobody can work out what's wrong with it, so the engineer walks around the machine, listens to it, examines a few things, then takes out a hammer and taps one particular spot. The machine immediately starts working again.
When he sends an invoice for $10,000, the owner is outraged. He was only there for a few minutes and all he did was hit the machine once with a hammer. So the engineer provides an itemised invoice: $1 for tapping the machine with the hammer, and $9,999 for knowing where to tap.
Whether that story ever actually happened isn't particularly important. The lesson is that the valuable part wasn't using the hammer. It was the decades of experience that allowed someone to know exactly where to use it.
AI is rapidly becoming an extraordinarily powerful hammer.
AI makes doing things easier
We've been using AI extensively in software development at Dcode. It can write code, analyse existing code, generate tests, help diagnose bugs, suggest database structures, build interfaces and dramatically accelerate repetitive development work. We're already seeing substantial improvements in what an experienced developer can produce when AI is used effectively.
Our clients are using it too. People who previously would have described themselves as non-technical are creating prototypes, analysing data, automating processes and building genuinely useful internal tools. I think this is one of the most exciting aspects of AI. The barrier between understanding a business problem and being able to do something about it has dropped enormously.
But there's an important distinction between being able to get AI to produce something and knowing what should actually be produced.
AI is very good at answering the question you ask. The harder problem is knowing what question you should be asking.
Experience is knowing what to ask
Imagine asking AI to "build me a job management system for my construction business." Today, AI can get surprisingly far with that instruction. It can create customers, jobs, tasks, forms, invoices and a nice-looking interface. You could have something running remarkably quickly.
But what should happen when a quote changes after a job has commenced? How should variations work? Should materials belong to the overall job or individual tasks? What happens when someone deletes a customer? What happens when two people edit the same job simultaneously? What happens when a worker is offline on a construction site and reconnects later?
Then there are questions around the underlying software. Who should be able to access sensitive attachments? What happens when the accounting integration fails halfway through a synchronisation? What needs to be logged? How are backups handled? How do you know those backups actually work? What happens when the business grows from five employees to 500?
AI can help answer every one of those questions. In many cases, it can suggest very good answers. But somebody needs to recognise that those questions need to be asked in the first place.
That's where experience becomes enormously valuable.
Domain expertise matters just as much
This was ultimately where the person I was debating on LinkedIn and I agreed. His 37 years in construction gives him knowledge about construction software that I simply don't have. He knows where the workflow actually breaks down, which administrative processes frustrate contractors and which features sound great in a software demonstration but won't survive five minutes on a real job site.
That knowledge is incredibly valuable, particularly now that AI gives him the ability to turn more of that knowledge into something tangible.
My experience brings something different. After more than 25 years of building and supporting software, I know the questions to ask about architecture, security, infrastructure, performance, maintainability, integrations, backups and recovery. I know that the decisions you make when a system has ten users can become very important when it has thousands.
Neither expertise makes the other redundant. In fact, AI potentially makes the combination far more powerful. Someone who deeply understands the problem can work with someone who deeply understands how to build and maintain the solution, with AI dramatically accelerating what both of them can accomplish.
AI doesn't eliminate expertise. It amplifies it.
I think we're sometimes looking at AI backwards. Much of the conversation is focused on which jobs AI will replace. I'm becoming more interested in what an expert can now accomplish with AI that they couldn't accomplish before.
Give someone with 30 years of construction experience access to AI and they can potentially build things they previously could only describe to a software company. Give an experienced software developer AI and they can potentially build in days what previously took weeks. Put those people together, give both of them AI, and suddenly you have deep domain expertise, deep technical expertise and an incredibly powerful execution tool working together.
That's where I think the real opportunity is.
As execution becomes easier and cheaper, judgement becomes more important. The ability to recognise the actual problem, ask the right questions, challenge an answer and know when something isn't right becomes a differentiator.
Knowing what's correct is the skill
We're discovering this every day using AI within our own development team. AI can confidently suggest an approach that looks completely reasonable. Sometimes it's excellent. Sometimes it's almost right. Sometimes it's completely wrong. My main concern is that it will never indicate this in a response - it is up to the user to interpret this and call out the response.
We've spent considerable time developing our own instructions, standards and AI skills around how we want software built. We're effectively trying to give AI the benefit of lessons we've learned over decades of building, maintaining and supporting software. We want it to understand not just how to produce code, but how we believe good software should be structured.
That doesn't mean our developers become less important. It changes where their value sits. Producing every line of code manually becomes less important, while understanding architecture, reviewing decisions, identifying problems and recognising whether AI's output is actually correct becomes more important.
And I don't think that's unique to software.
An accountant using AI still needs to understand accounting. An engineer using AI still needs to understand engineering. A construction professional using AI still needs to understand construction. AI can accelerate their work enormously, but it doesn't automatically give someone decades of experience or the judgement that comes from having seen the consequences of hundreds of previous decisions.
The opportunity isn't AI versus humans
I don't think the interesting discussion is whether AI will replace software developers, accountants, engineers or other professionals. The much more interesting question is how AI changes what experienced people in those professions are capable of producing.
We're going to see construction professionals building software. We're going to see accountants creating automation. We're going to see business owners analysing their own data and creating tools that would previously have required a development project. At the same time, we're going to see experienced software developers become dramatically more productive.
I think all of that is a good thing.
AI will supercharge our output. It will reduce the cost of experimentation, shorten the distance between an idea and a working solution, and allow people to create things they couldn't have created before.
But before asking AI to build something, there's still an important question to consider: Do I know enough about the problem to know what I should be asking it?
AI might be able to swing the hammer better and faster than any of us.
The value is still knowing where to tap.
