Coding was never the bottleneck

The allure of using AI tools to build software projects quickly has captivated the industry so much that many are preaching that we will no longer need ot hire software developers in a few years. Essentially, the notion relies on the fact that the AI models can write code faster than a human and makes the leap to conclude that there will eventually be no need to hire a human. However, the job of a software developers is not only to write code.

When I last attended the Render ATL tech conference, I recall listening to Kelsey Hightower explain that coding is a method of last resort to solving a problem. He emphasized that we are problem-solvers and should utilize whatever tool is at our disposal to implement the solution.

Coding is one of many tools that software developers use and, at least until recently, it was a very expensive tool to utilize in order to solve a problem. In other words, if a someone else has already created a solution, it is wasteful to spend time "recreating the wheel". That's why the entire industry has embraced using opensource.

A software developer's job is to understand the customer's problem so that we can quickly implement a solution using technology - whether or not they're writing the code.

Unfortunately, more and more of the industry are mandating the use of AI tools which often translates into a focus on writing more lines of code per minute. The business value of such a mandate is clearly to reduce the cost of software development. It was generally understood that an initial prototype (or MVP) may take 2-3 months of building. Now, we can prompt a LLM to build it in 2-3 days (and sometimes 2-3 hours).

I do understand the temptation. When we write the code to implement a custom solution, we are also essentially creating our own intellectual property. Since intellectual property has value then it makes sense to consider each line of code as an investment. Furthermore, given how rapidly the AI models are improving, the cost for making that investment is rapidly reducing - though some may argue otherwise. Regardless, the trend is clear and we can beat that these models will continue to improve.

However, that investment can easily become a burden - both in terms of the costs to maintain the chosen implementation as well as the cost to change direction after spending so much time implementing the solution ie the sunk-cost fallacy.

This burden (technical debt and the cost to change) is something that the industry generally understood - we stopped tracking lines of code as a means of measuring productivity and focused on delivering measurable value to the client. The Agile Manifesto encouraged us to focus on de-risking the process of software development by incrementally delivering value.

We've stopped thinking about whether or not we are sacrificing the fundamentals of software engineering at the altar of speed. As a result, when we're adopting these AI tools, we're fixing problems that either didn't exist before or that were already solved.

Moving fast in the wrong direction is worst than moving slowly

While the industry continues to obsess over the new capabilities of the latest AI model to write more production-ready code and whether or not the software development career is dying, many seem to have forgotten that there is no solution for not understanding the core problem to be solved. We can't replace first principles thinking.

Fortunately, there is growing interest on how we can utilize these tools to solve the real problems that we struggled with before such as identifying core customer problems which can be solved for a profit as well as figuring out mass-distribution of that solution. We need more of this.

If we don't understand the problem to be solved then we will still build the wrong thing....only faster.