By André van der Walt, Solutions & Software Engineer at Insight Consulting. Blog post 4 of 4 in the Prompted series.
The pros and cons of AI in the dev workflow are plentiful, and the impact is plain to see, good and bad. Code “slop” and fewer ways for juniors to grow are countered by never-before-seen levels of output and accessibility. It’s difficult to reflect on the negatives when the benefits are so abundant.
Unfortunately, keeping up with the industry affords little time to try to address the issues. The software industry feels like it is moving faster than it ever has, and one misstep leaves you in the dust. We haven’t solved anything, but we have some guidelines we follow as a team to help balance the need to hold pace and prevent some drawbacks where possible.
We treat AI as a “skilled junior”. It may seem disingenuous given the versatility and capabilities of AI, but that framing allows us to preserve a way of working that avoids delivering code we haven’t reviewed. We can define tasks, brainstorm, plan, research and implement in the same way you would with a junior, while guiding and overseeing the execution and validating the output.
Does this give us the most raw speed in output? No, but it does mean that we know exactly what enters our codebases and what we are delivering. When your business is grounded in client relationships, this matters. It is not to say that AI replaces juniors, especially not in the consulting context. The agency and ingenuity that come from the human mind are not quite replicated yet.
We have also changed how we do documentation. It has simultaneously become much faster to do and, counterintuitively, much more important. It captures functionality and intention when building using AI, for both human devs and other AI agents working in the project. It ensures an aligned vision for implementation, establishes codebase patterns, and avoids “drift” from the objective.
Beyond code, AI has found its place throughout our engagement with clients. It aids in research before meetings, requirements engineering and scope alignment, and drafting documentation. It also allows economical exploration of various avenues when solving the client’s request. Of course, this needs the same guardrails as code to ensure we know what we are presenting in a professional engagement. This is where we have seen the most benefit in the consulting and dev space. Being able to better explore and understand what to build, and more importantly, what not to build, is a massive boon.
On a more meta scale, we have also invested in knowledge sharing and internal tooling. We run short weekly sessions for the dev team to recap news in the AI space and share their experiences, experiments, victories, and where they got burnt. We built tooling that integrates with our selected AI tools to promote our internal ways of working, procedures and quality gates. It helps us all “sing from the same hymn book” while we implement or work on an engagement.
These practices have helped us stay on relatively firm footing while leveraging AI, but they come with their own set of caveats. Foremost in my mind is skill atrophy. With the ever-increasing utility of AI comes increased reliance. Where this is not recognised and reined in, it can quickly turn to over-reliance. When it becomes easier to spin the slot machine and get a usable output, you slowly forget the skills it is replacing. If we become nothing more than human interfaces to direct the output of the AI to the correct place, we are on a fast track to obsolescence.
However, this is not just a question of discipline. It is one of maintaining competitive relevance. If we opt not to use these tools to avoid that over-reliance, we end up being slower than our competition and thus commercially vulnerable. It’s a fine line we have been forced to walk.
Where atrophy worries me now, we face a larger issue in the future. Those of us already in positions deemed “senior” have skills that can atrophy. What about new skills? What about those who haven’t had the chance to learn at all? We treat AI as the junior, and in the process, we risk losing the ability to teach real juniors the lessons that build them up to senior positions.
But it doesn’t stop there. One of the core abilities of a good developer is to adapt and learn. AI has, to a large degree, removed this need because AI can produce good output without needing time to learn. This removes technical learning opportunities for new and seasoned devs alike and erodes the chance to develop the one thing AI does not have: good judgement.
AI has had an undeniably massive impact on our work. We are reaping the benefits and producing better work because of it, at record pace. But it would be folly to do so with blinkers on to the real risks that come with it. It serves no one if we lose the things that make us effective or create a craft inaccessible to new talent. I don’t have answers to these questions, so I offer them instead. How do we stay competitive, which means leaning on AI more, without losing the judgement we earned and honed with practice? Where does the next generation of judgement come from if the bottom rungs of the ladder have been removed?
This article first appeared in Insight Edge, our newsletter on LinkedIn.