In the year since we first launched our Industry Guide to AI (January 2024), many companies have evolved from learning the basics of the tools and techniques they needed to implementing, iterating, and improving on their implementation.

2024 marks the year that generative AI became a mission-critical imperative for the enterprise. The numbers tell a dramatic story: AI spending1 surged to $13.8 billion this year, more than 6x the $2.3 billion spent in 2023—a clear signal that enterprises are shifting from experimentation to execution, embedding AI at the core of their business strategies. This spike in spending reflects a wave of organizational optimism; 72% of decision-makers anticipate broader adoption of generative AI tools in the near future. This confidence isn’t just speculative—generative AI tools are already deeply embedded in the daily work of professionals, from programmers to healthcare providers.

Research from Menlo Ventures

While advancement from foundation models may slow, there is still an enormous amount of progress to be made to the speed, cost, and accuracy of GenAI inside your organization by adopting the best practices of peers and researchers.

RAG was the first example, as it was often the gateway for companies to begin experimenting with GenAI. There are now far more advanced and flexible styles of RAG, as well as tools and service providers who can help you to optimize your use of this technique.

Routers are another great example of how quickly the industry is changing and the benefits that are accruing to end users. Today, you can build your stack on top of a router that allows you to easily swap one model for another, shifting from private to open-source, third party to in-house, with minimal interruption to your GenAI functionality.

Agentic AI was around when we first published this guide, but it was largely being used by individuals hacking together personal projects—a wild west of an open-source community. AI agents have now gone mainstream, with companies like Anthropic, OpenAI, and Google offering agents that will take actions on a user’s behalf, controlling and interacting with various apps and services on their mobile device or desktop.

As progress on the pre-training stage of AI models has slowed, focus has shifted to adding more horsepower to the inference stage of the process. In the past, no matter how complex the query, most GenAI models aimed to deliver their response quickly. What the end user received was a sort of initial response—a first thought, if you will. Today, many systems allow users to specify if a complex problem should be routed to a system that takes more time to think, plan, test, and consider before responding. For use cases like basic customer service, this is probably not needed, and would add latency that could irritate customers. For users who are pursuing complex research and have no issue waiting minutes or even hours for high-quality answers, however, this new modality has the potential to deliver enormous value.

Here at Stack Overflow, 2024 brought some monumental changes to our business. We announced marquee partnerships for our data licensing business, built out our product offerings, and conducted research to substantiate the value our data can add to the performance of models fine-tuned on Stack Overflow data.

As the new year continues, we hope this refreshed version of our Industry Guide to AI helps to ground the most important developments happening in the GenAI space and offers practical information and advice that you can apply inside your organization.