As I finish building the first iteration of our in-house “agent platform” that is tailored to our specific needs, I found myself questioning if the criticism of the NIH(Not Invented Here) syndrome has changed as a result of AI-powered development.
Developing your own tools/libraries/software takes time, effort, money and creates a commitment for the future (maintenance, security). If readily available (and proven!) solutions already exist, it would often make sense to opt for those, even at a cost. The downside, of course, is that you often have to also spend time, effort, money to adapt them to your specific needs, or adapt our own tools to match how they work or what they can do. So I think it’s a delicate balance between the cost vs complexity.
But with the rise of AI-assisted development, and especially with the increase in Software Engineering capabilities of the newer models, I think the balance is shifting and perhaps we need to re-assess this equation.
Does it make sense to buy/download/learn/adapt a tool, when a capable AI model can do something tailored to your needs at an insignificant cost?
Of course the complexity/risk spectrum is very wide. One thing is to talk about a local tool to help parse CSVs and automate some excel sheets. Another is getting your company to use your own custom-built CRM that is publicly accessible in the internet.
I’ve been witnessing more and more AI-developed projects that respond to people’s real needs, and sometimes, to companies’ real needs and I think these are great.
The best clothes are the ones tailored to your body. Why wouldn’t software be the same? But while a custom-made suit is super expensive, we are probably soon going to reach a point where custom-made software is “cheap”.
This is my “rough and abstract” analysis on the NIH curve, considering the pre-ai times, now and what’s coming.