I’m going to skip the preamble of explaining why I’m writing this post. It’s not as if you haven’t already been impacted in some way by AI. Some industries have found it very useful, others have found it disruptive. But what about the web industry?
The Promise: Fully Automated Website Production
It’s a fact that software development costs time and money. The promise of AI was to reduce that time and automate a large amount of, if not all of, the work involved.
Software development was a good first target. Aside from the cost and skills shortage, it’s simple for computers to read code. And documentation is often freely available online for different languages, frameworks and tools.
But what has been the outcome?
The Reality: Software Rot
Like many tools, AI has proven to be useful. But it has limitations, some of which are difficult to spot initially. One of these is software rot.
You might think that software is good as long as it does what you need it to do. But this is only partly true: you find this out when you need to change what it does, or the way it does it.
Say for example you have built an ecommerce store. You’ve done it almost entirely using AI. You get a sense of elation as the first orders start to appear. However, you soon realise that it’s not calculating tax correctly.
You run it by the AI tool to fix it. You start to use the code it produces. But now you find something else has broken. You are no longer getting any orders through. This pattern repeats itself until you realise you need to call someone to fix it.
They likely tell you that the code needs some serious work if it is to perform it’s function correctly. The reason? The AI generated code contains contradictory, convoluted spaghetti code that is nigh on impossible to understand, let alone fix.
For this reason we have an optimistically cautious approach to using AI in our work.
The “Optimistically Cautious” Approach
We have found that AI is good for three things:
- Explaining documentation
- Exploring ideas
- Composing a first draft
In each of these scenarios we make sure the AI tool is given enough context to offer a reasonable solution. Without the proper context (for example, carefully structuring the query, supplying links to documentation, allowing access to the codebase in question), the AI will likely “hallucinate” solutions that are “wide of the mark”, or completely inappropriate.
1. Explaining documentation
If we are using an API or some other tool that either has badly written documentation or the documentation is vast and challenging to comprehend, we use AI to summarise it. This avoids needless wasted time.
2. Exploring ideas
Creative ideation is necessary taxing on the brain, which reduces the amount that can be achieved in one day. Sometimes we can use AI to explore or iterate on an idea, for example by supplying initial iconography and asking it to suggest several more designs based on the first.
3. Composing the first draft
If there’s a requirement to add some functionality to a site, we might ask the AI tool to suggest a solution. Sometimes it comes up with unexpected ideas that a human wouldn’t have thought about.
Whilst the result could include spurious, insecure or incorrect code, it allows us to see how the final product might look.
A Trained Eye
With each of these uses, it takes someone with experience to notice inaccuracies or convoluted code that will cause the code to rot. Experience that only comes from developing software or designing without AI.
As you can see, AI is useful. But it is one of the tools at our disposal, not the only one.
At each step of the way it’s the human interactions that drive business success. And only humans are capable of the kind of creativity that is needed for that success.