Re: How Ramp Builds Self-Maintaining Software
The new task for anyone working in the software business will be building the pipeline that will build software, deploy it, observe it, fix issues, add new features, in an endless loop!1 Everyone will have to become a tools maker; anyone who thought they could get away with just using tools will have to move one step further up and build the tools that they need.
We’re seeing this already with super customisable, minimal agents like Pi.dev. Maintaining the software that’s built on a base such as Pi is not easy. As it is very likely that the software built on Pi was itself written using AI, it will be a lot of code that no one really understands very much. This is the norm now.
Someone can start this process using this prompt:
Build a web application to handle payroll for my company which has 100 employees. Then, deploy it somewhere and give me the admin credentials. Also, add appropriate monitoring so that I know when it crashes or stops working.
They will have a running app in 4-12 hours, depending on how much involvement the operator wants to have with the code and its deployment. In order to get the loop going, you may use yet another long-running agent2 with the prompt:
Find bugs through the monitoring, and fix anything that you find. Keep the total cost under $50 per employee per month.
Set spending limits, hand your credit card to the AI agent, and let it do its thing. You will end up with something!
All of this is speculation. Some of this is possible already, with customized workflow, and there are stories of employees trying this out on the company dime. The only thing that is certain now is change. The software industry is going to change a lot in the next 10 years. All the predictions are going to be wildly wrong and we will end up in an end state that no one expected!