AI usage among programmers has only increased since its introduction. We began with a chat interface that we could query, craft program solutions, and copy + paste into our projects. Then we got AI assisted development patterns where the AI could sit in your editor and provide suggestions. We’re at a stage now where its more common to set an agent lose on a feature and let it autonomously run. At every stage the paved path is to use AI as a service models which implies sending and receiving messages over the internet.
That got me thinking — at what point will someone ask how to achieve all the niceties of AI agent programmers without having to make requests over the internet?
If I recall correctly local internetless development became a heuristic during the microservices boom. I wonder when it’ll become a similar heuristic for development with an AI?
If I were to make a prediction I suspect running AI models on one’s own hardware is going to become more common especially with open source models and workflows. AI programming should be treated as commodity and not necessarily a premium service. Hopefully as chip manufacturers perfect the silicon required to run models, they’ll become more practical to run on our own hardware!