AI Agents and the Business of Coding
It would seem that AI agents could eventually be highly efficient at producing and testing software code at scale.
Take Google (GOOGL) for example. Think of the software code Google has produced to develop and maintain its various products and services. Google’s code base is essentially a massive data set with which to train Gemini (Google’s Gen AI models), for the purpose of creating AI agents specifically tasked with code production (writing, testing, deployment of code). Platform companies such as Google and Microsoft (MSFT) are well-positioned to provide LLMs with unfettered access to massive code bases. Not every Software/technology company has this ability.
Software code production is the Gen AI use case that I am most bullish about. It is a well-defined use case. It is a use case with a clear problem (finite number of high quality software engineers) and a clear ROIC.
OpenAI has a head start in this area given that Sam Altman’s company has the first mover advantage in the Gen AI space.
Google is in a strong position given its massive code base, its proprietary TPU chips and LLMs (Gemini).
Microsoft ought to be in a strong position given its heritage as an application company and given its early mover position in the Gen AI space.
I believe that Amazon (AMZN) will be a strong player in the AI agent space as AWS pushes forward with Gen AI initiatives both in partnership with Anthropic and in terms of building its own LLMs. Aside from AI agents deployed for the purpose of Software code production, I believe that Amazon’s retail business will lean heavily on AI shopping agents over time.
Companies such as Atlassian (TEAM) play in the code management business and will likely think of creative ways to create and deploy AI coding agents.
The RPA companies excel at creating code that performs discrete tasks. That experience ought to carry over to developing AI agents.