George Lynch, Head of Technology Advisory at NashTech shares his thoughts on what generative AI mean for software developers. This article first appeared on ComputerWeekly.com.
This year has seen a remarkable explosion of interest in Generative AI. Go back just six months, and most people had never heard of it. Now, it seems as though everyone – from tech professionals to students – is using it and experimenting, in particular, with the ‘poster child’ ChatGPT.
This has also generated increasing discussion of whether AI and Gen AI spell the end for human roles, including in software development. Does ChatGPT and its Gen AI cousins (such as Bard, Copilot and Bing Chat) mean the demise of the software developer – given that Gen AI can already generate code snippets?
It occurred to me that a good first step in considering this question was… to ask ChatGPT. As always, it gave an instant, eloquent answer, and the thrust of its response was ‘no’. ChatGPT and similar language models, it wrote, “are unlikely to replace software engineers entirely”. This is because software engineering “involves much more than just natural language processing” and requires a range of skills such as problem-solving and collaboration that are “beyond the scope” of Gen AI.
However, it did also say that Gen AI “can automate certain aspects of software development” and “augment the capabilities of software engineers.”
Clearly, then, it’s not a case of ‘RIP software developer’ – or at least, not yet. But at the same time, there is no doubt that the advent of Gen AI will bring about significant changes and disruptions.
It’s helpful to divide the impacts into opportunities and risks. So, I’ll start firstly with the many positives.
Gen AI will significantly speed up certain aspects of the software development process. It can already manage some of the lower-level tasks, such as entry-level code writing, code snippets, testing and documentation. It can also write excellent comments in code – something software developers tend to be less enthusiastic about doing.
And that’s just as at today. These capabilities could (almost certainly, will) rapidly improve and increase going forward. What catapulted ChatGPT into the mainstream was its ability to leverage modern advances in computing power that allow Large Language Models (that underpin ChatGPT) to be trained in a relatively short period of time. The potential for Gen AI to leverage advances in quantum computing could open up even more opportunities for developers.
Who knows what agile will look like when Gen AI really develops? Imagine sprint cycles of not two weeks but two days or even two hours! The speed and productivity could be beyond our wildest dreams. The challenge will be handling all the code generated, curating it and managing it.
In short, there is massive potential to do things faster, potentially more cheaply, and to spend more of the human time involved on the higher-end, value-adding aspects. These could all be huge positives in terms of productivity and client delivery.
I don’t think there’s any question that senior developers and engineers will always be needed – they are where art meets science, they have the experience and the know-how and the creativity/problem-solving ability to bring it all together.
The role of the programmer and indeed business analyst will still be to elicit requirements from clients that can then be converted into prompts for GenAI to generate, test and document code snippets. These will still need to be woven into the fabric of the overall solution.
It’s also worth remembering that to get the most from Gen AI, you need well-constructed prompts. In a way prompting Gen AI is a protocol all of its own, akin to pseudo-code. Gen AI at the moment does not remove the need for thinking about how a software solution needs to be structured, which for the most part is the biggest value a software developer brings.
We must be honest and say that Gen AI does represent a threat to junior roles and entry level tasks. However, in a way this is no different to the new frameworks and automation tools that keep appearing in the market. It’s a factor the IT industry is already used to living with.
It may be more pronounced with Gen AI, however, as clients may expect software firms and consultants to reduce the size of their (human) teams due to the fact that they can use Gen AI – either that or get jobs done faster (or both).
An inflection point may therefore be coming. But tech has always been resilient and adaptable. It always reinvents itself. I have no doubt that new jobs and roles will be needed to support Gen AI (prompt engineers for example) that many junior team members can fill. Gen AI will be a disruptor, but I am confident that the industry will embrace Gen AI as it has other advances in computing science.
Whilst the industry will not only survive but thrive, the level of disruption Gen AI has caused and will continue to create may be a step too far for some digital leaders. They may decide to step back, allowing a new crop of leaders to step forward.
But there are other threats or risks that need to be managed, not just those around jobs.
There is arguably a danger of stagnation if Gen AI can only generate code similar to code used to train the underlying models. Will it ever be able to make innovative leaps forward? This is where human intelligence will likely always be needed and will remain at a premium.
Then there is the risk of error. There have already been cases of Gen AI being ‘confidently wrong’, suffering from ‘hallucinations’ caused by a lack of data, dirty data, or other constraints/errors in the models. Gen AI’s outputs, therefore, need to be checked, tested, validated etc – another area where humans are likely to keep their jobs.
It is essential that this happens. Otherwise, the speed at which false and/or malicious information (or code) could spread could be truly frightening and have potentially serious consequences.
Advice to digital leaders, consultants, practitioners
So where does this leave everyone in or connected to the IT industry? My five key pieces of advice are as follows.
As an example of how we are approaching it at NashTech, we’re actively trialling it and are about to start using a ChatGPT product, which we have developed and fine-tuned ourselves based on GPT4, for first line tech support for clients.
These are exciting times. No one can really be sure of the scale of the changes to come, but we can be sure that they’re going to be significant.
The advent of Gen AI has parallels with the advent of low-code/no-code in the last decade. Although low-code has replaced bespoke software development in many ways, there are actually more software developers now than a decade ago. This is because low-code/no-code, and also Gen AI, will always focus on the low hanging fruit of software development.
Once those fruits have been picked, forward thinking enterprises are keen to climb the tree further for the next innovation, and usually that requires bespoke work. How high is the tree and will Gen AI/low-code ever get to the top of it? We think very, very high, and no.
So, I don’t believe there is any danger of ‘RIP software developer’. Tech will remain a place where humans can thrive, building on the ever-more sophisticated outputs that AI brings us.
We've launched the 25th Anniversary edition of the Digital Leadership Report. The world's largest and longest running survey of senior technology decision makers.
Download the full report here.